perf: discard oversized query bookkeeping tables

[MichaReiser]

IndexMap clears retained hash control bytes in O(capacity), making sparse query (like 2 edges) frames expensive after a large query.

Ideally, this would be something that HashTable handles better. For now, we'll drop the allocation if we've seen one overly large IndexMap that is only very sparsely populated. Specifically, we preserve up to 2x slack above a 16K floor.

Testing: ty shows a 0.77% static-frame improvement.

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	0dcc897
🔍 Latest deploy log	https://app.netlify.com/projects/salsa-rs/deploys/6a6de86cb9d26000084996e9

[codspeed-hq]

Merging this PR will improve performance by 40.03%

⚡ 1 improved benchmark
✅ 73 untouched benchmarks
⏩ 1 skipped benchmark¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	WallTime	parallel_fast_path[NoEviction]	507.9 µs	362.7 µs	+40.03%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing MichaReiser:fix/discard-sparse-query-capacity (0dcc897) with master (59ab5b0)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩