tune tcmalloc for benchmarks

[anonrig]

Tcmalloc is extremely flaky while running benchmarks. This change would reduce the probability of outliers at the cost of diverging from production. For benchmarks that need tcmalloc to use the default values, they have to avoid including bench-tools.

Recommended by Codspeed folks.

[codspeed-hq]

Merging this PR will improve performance by 12.99%

⚡ 1 improved benchmark
✅ 67 untouched benchmarks
⏩ 129 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	simpleStringBody[Response]	21.6 µs	19.1 µs	+12.99%

---

_{Comparing yagiz/tune-tcmalloc (f328f8a) with main (4e603ca)}

Footnotes

1. 129 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[fhanau]

I think we'll need to select() based on //src/workerd/server:really_use_tcmalloc instead to handle the case of sanitizer builds (no tcmalloc but may be Linux) correctly. You may need to make that target public first for this to work.
Edit: Asan CI has passed – perhaps because we don't build benchmarks in the ASan job?

[anonrig]

Edit: Asan CI has passed – perhaps because we don't build benchmarks in the ASan job?

Probably yes. Regardless, I've added the changes you recommended.