fix: avoid quadratic synthesis line lookups

[danusha2345]

Summary

• replace the remaining per-match source.slice(0, index).split('\n') line calculations in dynamic-dispatch synthesis with the existing lazy newline index
• cover ArkUI, Gin, Redux thunk, object registries, Pinia, Vuex, Celery, Spring, MediatR, Sidekiq, Erlang, and Laravel passes
• pin both synthesized edge line metadata and the absence of the quadratic allocation pattern

Root cause and impact

Several regex-driven resolution passes recalculated a match's line by copying and splitting the entire source prefix. On a match-dense file, doing that once per match makes the final resolution pass quadratic on the main thread; reducing parse workers cannot affect it.

The existing makeLineAt helper builds one newline-offset array lazily per scanned source and resolves each match with a binary search. This keeps edge attribution unchanged while removing repeated prefix allocations.

This addresses one concrete main-thread amplifier relevant to the full-resolution OOM investigation in #1553. It does not claim that every reported 4–6.5 GB case has the same sole cause.

Validation

• npm run build:kernel
• npm run build
• npx vitest run --reporter=dot — 185 files passed; 3120 tests passed, 9 skipped
• affected synthesizer suites — 12 files, 28 tests passed
• controlled full-index A/B on a 20,000-match TypeScript fixture (three alternating runs): median 3.81s → 0.75s; median peak RSS 398,356 KB → 376,848 KB
• semantic graph hashes are identical before and after for nodes, edges, and all 20,001 unresolved references

[danusha2345]

Folded this exact commit into #1583 as 180c6ea so the quadratic resolution-pass work and the related memory fixes can be reviewed and tested together. Closing this draft as superseded; no code was dropped.