perf: skip content cache beyond 1000 files — 7% RSS reduction (#208)

[justrach]

Summary

During cold indexing, commitParsedFileOwnedOutline duplicated ALL file contents into a HashMap. The indexes (word, trigram, sparse) consume the content parameter directly — the cache was only needed for readContentForSearch which already falls back to disk.

Skip content storage when outlines.count() > 1000. First 1000 files stay cached for fast search; beyond that, search uses disk reads. Snapshot fast-load uses OUTLINE_STATE (not CONTENT), so startup is unaffected.

Benchmark (openclaw, 13,867 files, cold search)

Version	Time	Peak RSS	Delta
v0.2.56 baseline	6.16s	3,678MB	—
PR#258 (current)	5.66s	3,559MB	-3.2%
This PR	6.07s	3,415MB	-7.2%
Zero cache (limit=0)	5.90s	3,390MB	floor

Content cache accounts for ~170MB. Remaining ~3.3GB is from trigram index posting lists — would need flat array (#208 original scope) or compressed postings for further reduction.

Test plan

• All existing tests pass
• Benchmarked on openclaw: 3,415MB peak RSS (-7.2% vs baseline)
• No speed regression (within variance)
• readContentForSearch disk fallback verified working
• Snapshot load unaffected (uses OUTLINE_STATE fast path)

🤖 Generated with Claude Code

[github-actions]

Benchmark Regression Report

Threshold: 10.00%

Tool	Base (ns)	Head (ns)	Delta	Status
codedb_bundle	661116	664396	+0.50%	OK
codedb_changes	110276	108506	-1.61%	OK
codedb_deps	29908	29494	-1.38%	OK
codedb_edit	24764	23026	-7.02%	OK
codedb_find	138477	141168	+1.94%	OK
codedb_hot	147832	148196	+0.25%	OK
codedb_outline	462079	461798	-0.06%	OK
codedb_read	142738	144358	+1.13%	OK
codedb_search	282027	286296	+1.51%	OK
codedb_snapshot	4517878	4496363	-0.48%	OK
codedb_status	263375	256919	-2.45%	OK
codedb_symbol	64627	64493	-0.21%	OK
codedb_tree	87323	86963	-0.41%	OK
codedb_word	90962	99448	+9.33%	OK