perf: enlarge BoxedValues int cache to eliminate iterator index boxing

[rexm]

Summary

Follow-up to #652. Iterators box the per-item index through BoxedValues.Int on every iteration; indexes beyond the cache (previously 0–19) allocate a fresh 24-byte box each time. After #651/#652 removed the larger allocation sources, this was the dominant remaining allocation in list rendering — e.g. 23.5KB per render of a 1000-item {{#each}}, identical for object and dictionary data.

• Enlarges the cache from 20 to 1024 boxed integers (~32KB of process-lifetime statics), covering effectively all template loops.
• Also enables MemoryDiagnoser on the LargeArray and EndToEnd suites so allocation deltas are visible there (separate commit).

All 1840 tests pass.

Benchmarks

MediumRun (LaunchCount=1, 15 iterations), Apple M4, .NET 10. Baseline is current master @ 9ee2d48 (i.e. after #651/#652) plus the MemoryDiagnoser commit.

Benchmark	Case	master alloc	This PR	Δ time
RenderList	N=100, dictionary	1,920 B	0 B	+0.8%
RenderList	N=100, object	1,920 B	0 B	+0.0%
RenderList	N=1000, dictionary	23,520 B	0 B	+0.4%
RenderList	N=1000, object	23,520 B	0 B	+3.1%
RenderList	N=10, both	0 B	0 B	−6.2% / +0.7%
RenderToString	clean / html	31,656 / 34,960 B	30,936 / 34,240 B	−0.8% / +2.2%
LargeArray	N=20000	959,521 B	935,425 B	−1.0%
LargeArray	N=40000	1,919,521 B	1,895,425 B	−9.1%
LargeArray	N=80000	3,839,522 B	3,815,424 B	+2.3%
RenderNested	all 4 cases	unchanged	unchanged	−4.1% to +0.4%
EndToEnd	both	744 B	744 B	−3.0% / −1.5%

Time deltas are all within the ±3% cross-run noise band observed on this machine (the −9.1% and −6.2% outliers included) — this change is about allocation, not speed.

LargeArray drops exactly the 1,004 newly-cached index boxes (~24KB/render); its remaining megabytes are the List<int> values being boxed when read as object — a separate mechanism, out of scope here.

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud