[skia-sync] Merge upstream Skia main (tip)

[mattleibow]

Automated upstream merge of google/skia main (tip).

Upstream Skia merge — main tip into skiasharp (m150 tip sync)

This PR merges the bleeding-edge tip of google/skia main into our skiasharp branch.
This is NOT a milestone bump — the merged head branch is named skia-sync/main, our
user-facing milestone stays at m150, and only bug-fix / housekeeping deltas from upstream
flow in. (Note: upstream main's SkMilestone.h has already flipped to 151 — see "Items needing
human attention" below for how this PR handles that.)

Upstream merge details

Field	Value
Base branch (mono/skia)	skiasharp (was at b16789ec3540beaa43e8a71b050d9d328eb68025)
Upstream ref	google/skia main
Upstream tip merged	e793ca20f32abea8f96876d33c7ddabd3e5ed2d8 ("Temporarily disable sampling in SkPngRustCodec")
Merge base	ed03b304e16a27e3035175e9c90a5c3492b087e8
Commits ahead of merge base	~291 upstream commits
Merge commit	a2658b789bac600d1e49afa9f774e192b11dcb17
Post-merge fixup commit	cd67ad79830bdf6fd7f09f2e5328cf85aa06b10d
New skiasharp tip after sync	cd67ad79830bdf6fd7f09f2e5328cf85aa06b10d

Conflicts resolved

The upstream merge produced 7 conflicts. All resolved in the merge commit:

1. DEPS — auto-merge failed because of large refactors to infra_revision and several
   third-party pins. Resolved by taking HEAD's fork-specific content (pinned versions of brotli,
   expat, freetype, harfbuzz, libjpeg-turbo, libpng, libwebp, piex, vulkanmemoryallocator,
   spirv-cross, vulkan-headers, and the long list of # DISABLED: block comments) while
   accepting upstream's bumped infra_revision (604794226cd60e4225b0ee7a39f7944b8f280ec6).

2. bazel/external/dawn/dawn_files.bzl — upstream-deleted, locally-modified (UD). Upstream
   replaced the hand-maintained file list with a generator script
   (bazel/external/dawn/generate_dawn_files.py plus its test). Resolved by accepting the
   deletion — the older [m150] Update dawn_files.bzl after Dawn roll patch is no longer
   needed.

3. src/gpu/ganesh/ops/AtlasTextOp.cpp — three conflict hunks:

   • Includes (lines 14–22): upstream renamed include/private/base/SkOnce.h →
     include/private/SkOnce.h etc. Took upstream's new header paths.
   • AtlasTextOp::onCombineIfPossible overflow logic (lines 188–192): upstream cherry-picked
     our [ganesh] Avoid overflow when combining AtlasTextOps change (eb4f33fe2b). Took
     upstream's identical refined form.
   • Stride mismatch handling (lines 579–592): we had a verbose SKIA_LOG_D + early-return.
     Took upstream's single-line SkASSERTF_RELEASE(strideCheck == vertexStride, "stride mismatch"); — converges on the upstream policy of treating this as a bug rather than a
     silently-swallowed runtime condition.

4. src/gpu/ganesh/ops/StrokeTessellateOp.cpp — include rename. Took upstream
   (src/core/SkArenaAlloc.h, src/core/SkSafeMath.h). Our M150 cherry-pick
   [M150] Prevent overflow in StrokeTessellateOp (fd9c3a7681) already made it upstream as
   dcf3ebf072.

5. src/gpu/graphite/dawn/DawnGraphicsPipeline.cpp — include rename to
   include/private/SkLog.h, include/private/SkTArray.h, src/core/SkTBlockList.h. Took
   upstream.

6. src/gpu/graphite/render/SDFTextLCDRenderStep.cpp — two hunks, both due to the upstream
   include refactor + the conditional compilation guard #if !defined(SK_DISABLE_SDF_TEXT). Took
   upstream. Our M150 cherry-pick [M150] Turn off LCD in SDF slugs (86bafa2700) made it
   upstream as 95dbfa24e0 (refined to use sk_ieee_float_divide(1.f, getMinScale()) instead of
   our getMaxScale() formulation — accept upstream's correction).

7. src/ports/SkTypeface_mac_ct.h — include rename to include/private/SkOnce.h +
   src/core/SkSharedMutex.h. Took upstream. Our M150 cherry-pick
   [m150] Resolved a Data Race on fStream in SkTypeface_Mac (14d05ec761) made it upstream as
   ba3ee9b265.

In short, every M150 cherry-pick we previously carried on the skiasharp branch has been
upstreamed. The merge drops those private patches in favour of the canonical upstream forms.

Major upstream changes pulled in

• Massive header refactor — include/private/base/Sk*.h → include/private/Sk*.h and several
  src/base/Sk*.h → src/core/Sk*.h (or include/private/). Affected ~116 files (visible as R
  entries in git status). The only code we own that referenced the old paths was
  src/c/sk_font.cpp (see C API fixes below).
• partition_alloc integration — new directory src/partition_alloc/ with a noop/ fallback
  for builds that disable PartitionAlloc. New gn arg skia_use_partition_alloc (default
  is_clang) controls this. See "Items needing human attention".
• Sparse strips for Graphite — new src/gpu/graphite/sparse_strips/ directory
  (MakeStrips.h, StripProcessorScalar.h).
• VulkanGraphiteUtils.h removal from include/gpu/graphite/vk/ — upstream consolidated
  the public surface (no SkiaSharp impact — we don't bind Graphite Vulkan utils).
• Dawn rolls + ANGLE rolls + vulkan-deps roll + Skia infra roll — captured in the merge.

C API fixes in cd67ad7983

Two minimal edits were needed in our shim to compile against the new tree:

1. include/core/SkMilestone.h: pin SK_MILESTONE back to 150. Upstream
   chrome-branch-day flipped it to 151 (commit d30d2a1958) on main. We do not bump our
   user-facing milestone in tip-mode syncs, so we pin the underlying constant back to match
   scripts/VERSIONS.txt. The Cake build's git-sync-deps task asserts these are in sync —
   without this pin the build aborts before compilation. (Future option: bump VERSIONS.txt and
   drop this pin once we're ready to formally move the SkiaSharp line to m151.)
2. src/c/sk_font.cpp: #include "include/private/base/SkTemplates.h" →
   #include "include/private/SkTemplates.h" after the upstream header move.

Items needing human attention

• Upstream milestone has already moved to 151. SkMilestone.h was bumped on
  main as part of the normal chrome branch day flow. This PR keeps SK_MILESTONE = 150 to
  match SkiaSharp's pinned milestone in scripts/VERSIONS.txt. A reviewer should confirm:
  • (a) This is the intended behaviour for a tip-mode sync that crosses a chrome-branch-day, OR
  • (b) We should instead promote SkiaSharp to m151 in a follow-up (bump VERSIONS.txt milestone,
    cgmanifest chrome_milestone/version, soname, all NuGet version lines) and drop the
    SkMilestone.h pin.
• AtlasTextOp.cpp stride-mismatch change is a behaviour change: we previously logged and
  bailed out; upstream now releases an assertion. In debug builds this is fine; in release/CI
  builds this could surface as a hard crash if the condition ever fires. Recommend a quick sanity
  pass over any internal/runtime users that may rely on the old "log-and-skip" recovery.
• partition_alloc is intentionally disabled in our build via the new --gnArgs "skia_use_partition_alloc=false" argument passed at dotnet cake --target=externals-linux
  time (this is a command-line gn arg, not a native/**/build.cake modification). When this
  sync eventually graduates into a real milestone bump, we should consider whether to wire
  skia_use_partition_alloc=false into all platform build.cake files (single source of truth)
  rather than relying on the CLI argument every time. Filing as a follow-up.
• bazel/external/dawn/dawn_files.bzl deletion accepted. Anyone currently relying on the
  hand-maintained file list should switch to the generated form.
• DEPS infra_revision advanced to 604794226cd60e4225b0ee7a39f7944b8f280ec6 while
  keeping all fork-specific dependency pins. Confirm this is the right thing for SkiaSharp's
  infra/CI tooling (this only affects Skia's own bots, not our build).

Created by skia-upstream-sync.

[mattleibow]

Closing stale skia-sync validation branch/PR to give the refactored auto-skia-sync workflow a clean slate for a fresh run.