L3 WALA engine: deferred minor follow-ups from the #194 review #202

Is your feature request related to a problem? Please describe.

Follow-ups deferred from the #194 L3 WALA engine review (merged in PR #196) — tracked so they don't get dropped. All are non-correctness; the merged engine is sound. Flagged by the final whole-branch review as follow-ups.

• Skip-counter granularity (L3WalaOverlays): skippedNoMatch conflates three distinct coverage-skip causes — type-not-in-index, signature-not-in-type, and no source BlockStmt. Split into per-cause counters and surface them individually in the coverage log so low-coverage debugging is actionable. (review item M1)
• PDG control-dependence test (WalaPdgBuilderTest.controlDepEdgeExistsFromIfToGuardedStatement): asserts only the edge source (the if-test line); also assert the destination (the guarded statement), so it proves the branch controls its guarded statement rather than merely that an edge leaves the branch. (review item M3)
• do/while degenerate fallback (WalaCfgBuilder.wireConditional): the takenBlock == null fallback path has no test coverage. Add a fixture that exercises it.
• (Optional) phi-mediated scalar DDG: WALA's NORMAL→NORMAL filter drops branch-merge / loop-carried scalar def-use pairs (a phi is not a NormalStatement); the heap points-to DDG is sound. Resolving through phi/pi nodes would bring scalar-DDG completeness closer to the AST engine. See docs/design/notes/l3-dataflow-metrics.md ("phi-mediated-scalar limitation").

Part of epic codellm-devkit/.github#42.

Describe the solution you'd like

Not stated in the original issue.

Describe alternatives you've considered

Not stated in the original issue.