fix: LRU edge cases in TTL handling and factory validation

[avoidwork]

Description

Fixes 14 edge cases in TTL handling and factory validation for the LRU cache. Validates constructor inputs, reclaims expired keys on set(), skips expired items in read methods, makes forEach() mutation-safe, fixes get() stats, validates array inputs, corrects expiry=0 semantics, and silences setWithEvicted() eviction notification.

Type of Change

• Bugfix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Refactor (no functional changes)
• Performance improvement
• CI / build / tooling

Testing

Added 19 new unit tests covering the edge cases: factory validation with non-integer/negative/boolean inputs, expired-key set() with both resetTTL values, read-method TTL enforcement, forEach() mutation safety, get() stats on expired, getMany()/has*() array validation, expiry=0 semantics, and setWithEvicted() silent eviction. Updated 3 existing tests for the corrected expiry=0 behavior.

Coverage

• 100% line coverage maintained

Checklist

• npm run lint passes
• Tests pass with 100% line coverage
• No forbidden patterns used
• Conventional Commit style applied

[avoidwork]

Performance Audit — No Regression

Ran the benchmark:modern suite as an A/B against the pre-PR build. Verdict: net-neutral to slightly positive. Safe to merge.

Results (pre-PR vs. PR)

Improvements (the bulk):

Benchmark	Latency Δ	Throughput Δ
get-hit-sequential-100	-17.0%	+20.5%
set-random-empty-cache-1000	-11.8%	+1.4%
get-hit-random-1000	-8.1%	+9.4%
set-random-full-cache-100	-8.3%	+5.3%
get-miss-5000	-8.9%	+1.0%
has-operation	-6.8%	+1.7%

Slight regressions (all within noise):

Benchmark	Latency Δ	Throughput Δ
get-hit-random-5000	+4.2%	-3.1%
get-hit-sequential-5000	+4.6%	-3.3%
object-values	+9.8%	-1.9%
set-sequential-empty-cache-5000	+4.6%	+1.4%

Why the regressions aren't real

The 5000-size cases report ±19% variance in the benchmark's own output. A +4% shift on a ±19% measurement is pure noise. The object-values +9.8% is a single-sample blip on a 65ns operation — also noise.

Code-level reasoning

• #removeItem / #evictItem extract duplicated logic — no new work, just consolidation. delete + #unlink + pointer-null is the same work as before.
• #isExpired now does one fewer comparison (item.expiry === 0 removed) — strictly cheaper on the hot get/set/has path.
• The added #isExpired branches in values/entries/forEach/toJSON add a per-item check, but those aren't benchmarked hot paths, and the branch is cheap (one comparison).

Bottom line: The get/set/has hot paths are actually faster, and the only "regressions" are inside the noise floor. Safe to merge.