perf(mpt): specialize remaining common RLP headers

[Qumeric]

Stacked on #666.

Handle canonical short-string headers and two-byte long-list lengths directly during MPT witness decoding. The dominant 0x80 and 0xa0 paths remain first and unchanged, while uncommon forms retain the generic fallback.

Benchmark results

Block 24001988, RVR execution: #666, this PR.

metric	#666	this PR	delta
instructions	635,001,371	632,797,192	−2,204,179 (−0.35%)
rows (unpadded)	918,403,605	916,168,884	−2,234,721 (−0.24%)
main cells (unpadded)	46,698,352,107	46,640,829,085	−57,523,022 (−0.12%)
interaction cells (unpadded)	14,296,765,843	14,264,136,920	−32,628,923 (−0.23%)

Block 24003902

RVR execution: #666, this PR.

metric	#666	this PR	delta
execute_metered_insns	510,039,370	509,497,798	−541,572 (−0.11%)
metered_rows_unpadded	690,800,979	690,257,462	−543,517 (−0.08%)
metered_main_cells_unpadded	30,015,457,221	29,999,653,205	−15,804,016 (−0.05%)
metered_interaction_cells_unpadded	11,954,965,713	11,947,014,482	−7,951,231 (−0.07%)
metered_memory_unpadded_bytes	573,582,845,194	573,403,752,933	−179,092,261 (−0.03%)
memory-triggered segment cuts	51	51	0

[Qumeric]

Collapsed into #684 together with the other two RLP PRs. They were one idea split across three stacked PRs; #684 carries the union plus an rlp module and differential tests against alloy_rlp.