#9106: Add PDiff similarity score to Version Tracking

[clearbluejar]

Summary

Add a PDiff (basic-block mnemonic hash) similarity score to the Version Tracking match
table. The score is computed at match creation time, persisted in the database, and
exposed via a new Similarity column and filter in the UI.

Related: #9106, #5859

Problem

Version Tracking correlators each produce their own similarity scores, but these reflect
each correlator's matching algorithm rather than actual structural similarity. For example,
the Symbol Name correlator assigns a perfect 1.0 when names match, even if the functions
differ significantly. There is no correlator-independent metric for how structurally
similar two matched functions are at the basic-block level, making it difficult to
prioritize matches when patch diffing.

Solution

New PDiff similarity score — computed once at match creation time and stored in the DB.

Score formula

• 95% basic-block mnemonic hash similarity — for each basic block, mnemonic hashes are
  collected, sorted (to tolerate compiler instruction reordering), and combined into a
  per-block hash. A sorted-merge counts matching blocks between source and destination.
• 5% stack frame size similarity — min(a,b)/max(a,b) ratio; a subtle tiebreaker that
  distinguishes otherwise-identical functions with different local variable allocation.

New UI components

• Similarity column — displays the stored PDiff score in the match table
• Similarity filter — filter matches by PDiff score range (bypasses DATA and null-score
  matches)
• Best PDiff Match filter — deduplicates matches across correlators, keeping the best
  score per source/destination function pair

Backward compatibility

• Auto-upgrade: v0 sessions are automatically migrated to v1 schema on open
• Backfill: After programs are loaded, any function match with a null PDiff score is
  automatically computed and persisted — upgraded sessions get full scores on first open
• Filter safety: Null scores (DATA matches, pre-migration) pass through the Similarity
  filter instead of being rejected

Changes (15 files, 846 insertions, 13 deletions)

Core DB schema

• VTMatchTableDBAdapter.java — Add PDIFF_SIMILARITY_SCORE_COL, bump schema 0→1
• VTMatchTableDBAdapterV0.java — Accept v0 and v1, auto-upgrade v0→v1, write new column
• VTMatchDB.java — Read/write stored PDiff score
• VTSessionDB.java — Wrap adapter init in transaction (for upgrade), backfill on program open

Domain model

• VTMatch.java — Add getPdiffSimilarityScore() interface method
• VTMatchInfo.java — Add pdiffSimilarityScore field with getter/setter

Score computation

• VTMatchSetDB.java — Compute PDiff score in addMatch() for FUNCTION matches
• BasicBlockMnemonicFunctionBulker.java — Per-basic-block mnemonic hashing + combined
  similarity
• FunctionBulker.java — Interface for function hash strategies

UI components

• AbstractVTMatchTableModel.java — New Similarity column reading stored value
• SimilarityFilter.java — New filter by PDiff score range
• BestPDiffMatchFilter.java — New deduplication filter across correlators
• VTMatchTableModel.java — Register Similarity column
• VTMatchTableProvider.java — Register Similarity and BestPDiff filters

Support

• MatchMapper.java — Delegate getPdiffSimilarityScore() for implied matches

Test plan

• All 6 existing VT database tests pass (./gradlew :VersionTracking:test)
• Build compiles clean (./gradlew jar)
• New VT sessions compute and display Similarity scores immediately
• Old v0 sessions auto-upgrade and backfill scores on open
• Similarity filter works responsively
• DATA matches and null-score matches pass through filter correctly
• BestPDiff filter correctly deduplicates across correlators