Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@pkg/meta/model/table.go`:
- Around line 584-596: Resolve the merge conflict in table.go by removing the
Git conflict markers and keeping the incoming GetIdxChangingFieldType helper.
Ensure the table package compiles by deleting the leftover <<<<<<<, =======, and
>>>>>>> lines while preserving the function and its comment, and verify no other
conflict markers remain in this area.

In `@pkg/planner/core/expression_rewriter.go`:
- Around line 380-387: Resolve the merge conflict in the expressionRewriter
struct by removing all conflict markers and keeping the incoming fields
astNodeStack and useNewCollate alongside planCtx, since expressionRewriter
methods such as the ones referencing er.useNewCollate depend on that field to
compile. Make sure the struct definition is consistent with the rest of
expressionRewriter in expression_rewriter.go and that no HEAD-only version of
the struct remains.

In `@pkg/table/tables/index.go`:
- Around line 32-57: Resolve the remaining merge conflict markers in the
package-level declarations and imports in index.go so the file can compile
again. Remove all raw conflict artifacts like the HEAD/ab1e19714d6 sections and
reconcile the duplicated imports and declarations around indexConditionECtx and
indexPartialCondition, keeping the intended definitions from the merged changes
in this file. Verify any affected symbols in the table/index logic still
reference the same types and helpers after the conflict cleanup.
- Line 46: The global index parsing context is left as a zero-value
BuildContext, which can panic when ParseSimpleExpr emits warnings during
partial-index condition parsing. Initialize indexConditionECtx with a valid
exprctx.BuildContext before it is used by the index condition parsing path in
tables/index.go, and make sure the ConditionExprString handling always passes a
non-nil evaluation context through ParseSimpleExpr and ctx.GetEvalCtx().

In `@pkg/table/tables/tables.go`:
- Around line 270-277: Resolve the merge conflict markers in initTableIndices by
removing the <<<<<<<, =======, and >>>>>>> text and keeping the
NewIndexWithCollate(t.encoder.UseNewCollate(), t.physicalTableID, tblInfo,
idxInfo) path with its err check and early return. Make sure the resolved code
in tables.go uses the NewIndexWithCollate call instead of NewIndex so the index
creation remains explicit about collation handling.

In `@pkg/table/tables/testutil/indexcheck.go`:
- Around line 38-48: The index lookup in the test helper leaves idx nil when
indexName is not found, and the later idx.Meta() call will panic instead of
reporting a useful test failure. In indexcheck.go, update the lookup around
tbl.Indices() to explicitly assert that a matching index was found before
calling tablecodec.GenIndexKey, using the existing require-based test flow in
this helper so the failure is actionable and tied to the indexName/table.Index
search.

In `@pkg/tablecodec/tablecodec_test.go`:
- Around line 751-752: The test file contains unresolved merge-conflict markers
that prevent compilation; remove the conflict markers from the affected test
sections and keep the incoming cherry-pick side’s additions. Update the
`tablecodec_test.go` test blocks around the new test functions so only the
intended test code remains, and also clean up the same merge-conflict marker
occurrence in the other referenced section.
- Around line 888-889: The test assertion in tablecodec_test.go is using
require.NotContains on key with a []byte value, which makes the check
ineffective. Update the unique index key assertion in the relevant test case to
compare against PartitionIDFlag directly, or switch to the same byte-scan style
used for the positive check below so the absence of the flag is actually
validated.

In `@pkg/tablecodec/tablecodec.go`:
- Around line 1621-1625: Resolve the merge conflict in the tablecodec
restoration check by removing the conflict markers and keeping the incoming
collation-aware path in the relevant tablecodec logic; update the condition that
currently references types.NeedRestoredData and use the version that calls
types.NeedRestoredDataWithCollate with model.GetIdxChangingFieldType(idxCol,
col) and useNewCollate so the collation flag is threaded through explicitly.

---

Nitpick comments:
In `@pkg/table/tblctx/buffers.go`:
- Around line 90-97: EncodeBinlogRowData still creates its own encoder with
collate.NewCollationEnabled(), which leaves an implicit global dependency and
can disagree with the caller’s collation context. Update EncodeBinlogRowData to
accept a codec.Encoder parameter, and thread that encoder through to
tablecodec.EncodeOldRow instead of constructing it locally, matching the
explicit encoder flow used by WriteMemBufferEncoded and the EncodeRowBuffer
type.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@pkg/infoschema/issyncer/loader.go`:
- Around line 293-303: Add unit coverage for the new filter-driven decision
paths in Loader: create a fake Filter in loader_test.go that exercises
skipLoadingDiff when SkipLoadDiff returns true and fetchAllSchemasWithTables
when IterDatabases/SkipLoadSchema causes schemas to be skipped. Also add a brief
intent comment above the new else if l.filter != nil branch in Loader to explain
it is used by the lightweight bootstrap domain to load only selected schemas.

In `@pkg/meta/reader.go`:
- Line 32: The IterDatabases method in the reader interface leaves its callback
parameter unnamed, which is inconsistent with IterTables. Update the
IterDatabases signature to name the func argument using the same local
convention as IterTables (for example, fn) so the interface stays stylistically
consistent with nearby code.