perf(evm): fold edge coverage optimizations

[decofe]

Folds the edge-coverage-related draft perf PRs into one reviewable PR:

• perf: inline edge coverage hooks #15257
• perf: preallocate dense edge coverage map #15260
• perf(evm): use entry for cmp site counts #15291
• perf(evm): compact edge coverage depth #15293
• perf(evm): compact cmp site pc key #15294
• perf(evm): defer edge coverage depth lookup #15295
• perf(evm): insert first dense edge hit directly #15302
• perf(evm): cache repeated cmp log site #15312

The folded changes are limited to edge coverage collection and cmp-log bookkeeping:

• pass already-loaded opcodes into edge/cmp helpers and inline hot helpers
• pre-size dense edge maps for collision-free coverage
• compact optional call-depth storage
• use entry APIs for cmp site and dense edge counts
• cache consecutive cmp-log sites

Validation:

• cargo fmt --all
• cargo test -p foundry-evm edge_cov --lib

Prompted by: @grandizzy

[mablr]

I benchmarked the edge-coverage changes against origin/master with temporary targeted microbench tests inside the affected edge_cov.rs module, built with Foundry's profiling profile. I avoided full scfuzzbench here because a short campaign was noisy and dominated by setup/build work; these microbenches isolate the changed hot paths directly.

Results:

Benchmark	master	this PR	delta
repeated cmp-site logging	8.526 ns/op	8.353 ns/op	~2.0% faster
dense edge hits	15.564 ns/op	15.585 ns/op	neutral

Recommendation: merge, but I would describe the performance impact as modest. The repeated cmp-site path shows a small targeted win, while dense edge storage is effectively neutral in this synthetic benchmark. The implementation still looks directionally reasonable for a hot path.