perf(mpt): specialize common RLP headers

[Qumeric]

Stacked on #665.

Specialize the canonical RLP headers used by nearly every resolved MPT node: short list headers, one-byte long-list headers, and fixed 32-byte child references. Uncommon and malformed forms retain the generic alloy_rlp decoder as a checked fallback.

Benchmark results

Block 24001988, prove-stark, RVR, g7e.2xlarge: before, with specialized headers.

metric	before	with this PR	delta
execute_metered_insns	640,063,687	635,122,863	−4,940,824 (−0.77%)
metered_rows_unpadded	923,404,883	918,512,602	−4,892,281 (−0.53%)
metered_main_cells_unpadded	46,872,486,323	46,712,135,069	−160,351,254 (−0.34%)
metered_interaction_cells_unpadded	14,381,062,111	14,298,657,640	−82,404,471 (−0.57%)
total_cells	63,193,141,279	62,961,656,223	−231,485,056 (−0.37%)

Block 24003902

RVR execution: #665, this PR.

metric	#665	this PR	delta
execute_metered_insns	511,200,060	510,039,370	−1,160,690 (−0.23%)
metered_rows_unpadded	691,954,121	690,800,979	−1,153,142 (−0.17%)
metered_main_cells_unpadded	30,054,639,856	30,015,457,221	−39,182,635 (−0.13%)
metered_interaction_cells_unpadded	11,974,365,720	11,954,965,713	−19,400,007 (−0.16%)
metered_memory_unpadded_bytes	574,055,559,407	573,582,845,194	−472,714,213 (−0.08%)
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.