perf(evm): fold corpus mutation optimizations

[decofe]

Folds the small corpus mutation perf PRs into one reviewable PR:

• perf(corpus): skip all-empty cmp payloads #15300
• perf(evm): avoid single arg mutation index alloc #15310

The folded changes are limited to corpus mutation bookkeeping:

• avoid storing all-empty cmp payload vectors
• avoid allocating a one-element mutation index vector for single-arg ABI mutation

Validation:

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

Prompted by: @grandizzy

[mablr]

I benchmarked the corpus changes against origin/master with temporary targeted microbench tests inside corpus.rs, built with Foundry's profiling profile. The benchmarks cover both changed paths: single-argument ABI mutation and all-empty cmp-sequence handling.

Results:

Benchmark	master	this PR	delta
single-arg abi_mutate	446.471 ns/op	411.623 ns/op	~7.8% faster
all-empty cmp-seq storage, 10k inputs	48.504 ns/input, stores 10k empty cmp entries	146.283 ns/input, stores 0 cmp entries	CPU slower, payload much smaller

Recommendation: merge, but frame the cmp-sequence change as a memory/storage win rather than a CPU win. The single-argument mutation optimization is a clear targeted CPU improvement. The empty cmp-sequence path spends more CPU to detect and trim the all-empty payload, but avoids retaining/persisting thousands of empty cmp vectors, which is likely the more important behavior for corpus size and serialization overhead.