fix: switch scanJsonlFile and parseSessionFile to readSessionLines to prevent OOM

[maucher]

Fixes #131

Problem

readViaStream (the code path for files ≥ 8 MB introduced in #67) reassembles the entire file into a single string via chunks.join('\n'), giving the same peak allocation as a plain readFile. Callers then do content.split('\n'), creating a second full copy. With FILE_READ_CONCURRENCY = 16 and files up to 128 MB, theoretical peak heap usage is ~6 GB — right where the crash lands:

FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory
...
v8::internal::JsonParser<unsigned short>::ParseJson

Fix

readSessionLines already exists in src/fs-utils.ts as a proper async generator that yields one line at a time and never holds the full file in memory. This PR switches the two hot-path callers to iterate it directly:

• scanJsonlFile in src/optimize.ts
• parseSessionFile in src/parser.ts

No concurrency change needed — with true line-by-line streaming, 16 concurrent files each hold only one line at a time.

Tests

Two new tests added to tests/optimize-fs.test.ts:

• Spy test — confirms readSessionLines is called and readSessionFile is not called by scanJsonlFile
• 500-entry correctness test — verifies all entries in a large multi-line file are processed without truncation

Checklist

• npm run build passes
• npm test passes (323 tests, 0 failures)
• No Claude/Anthropic co-author trailers in commits