feat(mixed-filament): phase 2 hardening — ratio cap, recipe dedupe, cancel-aware rematch

[zhangzhend0ng]

Description

Phase 2 hardening of the mixed-color batch-match flow. Builds on the phase-1
batch dialog with four areas of work: a per-component ratio cap, duplicate-
recipe deduplication, cancel-aware re-match, and UX polish.

What's new

Per-component ratio cap (max_component_percent)
Threaded through the whole match chain (build_best_color_match_recipe →
batch_match_model_colors → recommend_best_filament_combo), limiting any
single filament's share of a mix:

• RECOMMENDED mode: [0%, 70%] — product spec: no single component > 70%.
• MANUAL mode: [15%, 100%] — near-pure allowed; over-70% surfaces as a
  post-match advisory via check_manual_recipe_ratio(), listing all
  over-threshold filament IDs.

Duplicate-recipe deduplication (merge_duplicate_recipe_mappings)
Mappings whose non-pure recipes are byte-identical (same components + ratio +
gradient) now share one virtual slot instead of each allocating a duplicate
mixed-filament row. The survivor unions source_extruder_ids and
merged_model_indices; runs after both ID-assignment phases so it keeps the
final target_filament_id. Pure recipes pass through untouched.

Cancel-aware re-match
recommend_best_filament_combo returns {} for both "no combo" and
"cancelled" — disambiguated via a post-call cancel_token re-check, routing
a cancelled result (error_code 2) directly with no error banner and skipping
the Pass-2 match. On a failed/cancelled re-match, the prior preview is
restored only when user-facing input is unchanged (mode + manual selections).

"CMYW" → "Full Spectrum" rename
New canonical constant kFullSpectrumPresetName; slots 1-4 auto-assigned the
preset after batch match so the combobox shows the preset name, not F1-F4.

Bug & safety fixes

• Dangling reference (UB) in extract_model_colors: ternary bound a
  temporary to a reference (use-after-scope). Now a raw pointer with null-check
  at the virtual-lookup site; null warning logged once (was per-face).
• Descending-invariant hardening in cleanup_unused_filaments_after_batch_match:
  promoted from assert() (compiled out under NDEBUG) to a runtime check
  with a per-deletion fallback that keeps colours correct on violation.
• mix_b_percent clamped to [0,100]; 64-color warning deferred until after
  build_ui(); preset_bundle null-guarded at startup.
• Triple-loop wb upper bound now respects max_component_percent.
• C-style casts → static_cast; m_match_running documented UI-thread-only.

UX polish

• Recommended-card hover tooltip: localized color name + hex + TD (resolved by
  color family, not palette position).
• Row border wraps swatch + name together; banners moved below the mode row.
• canvas3D() → get_view3D_canvas3D() on the thumbnail render path.
• Skip refresh on method re-select (wxMSW combobox quirk).

Tests

• Build: cmake --build build --target Snapmaker_Orca --config Release
• Recommended-mode match: no single component > 70%
• Manual-mode match with near-pure selection: over-threshold advisory lists all offending IDs
• After confirm, combobox shows Full Spectrum preset names (not F1-F4)
• Two models whose recipes collide share one virtual slot (no duplicate row)
• Cancel a running match: no error banner, prior preview restored when input unchanged
• Regression: batch-match cleanup produces correct colours when redundant filaments removed
• Regression: no crash on early-startup paths where preset_bundle is null

[Kenshin627]

for循环外层判断?

[zhangzhend0ng]

为什么不能直接"提到循环外"
pb 只在虚拟混合丝分支（idx >= filament_colours.size()，line 1334）里需要。物理颜色分支（line 1331-1333）完全不依赖 pb：

cpp
if (idx < filament_colours.size()) {
color_hex = filament_colours[idx]; // 物理颜色 — 不需要 pb
} else {
if (pb == nullptr) { continue; } // 虚拟颜色 — 需要 pb
...
}
如果把 if (pb == nullptr) 提到最外层循环并 continue/return，会把物理颜色也一起跳过——那是错的。所以这个 null 检查必须留在 else 分支里，它检查的是"这次访问需不需要 pb"，不是"整个循环要不要跑"。