perf(l1): improve rlp decoding

[edg-l]

Motivation

Basic integers like u64 used static_left_pad, that uses a temporary zeroed array, now they directly produce the integer. Did this by using decode_be_integer_u64, whose result can be casted to the relevant iX and decode_be_integer_u128, separate to avoid paying the cost of u128 always.

Modified decode_rlp_item so that it doesnt use usize::from_be_bytes(static_left_pad(x)), and instead uses a dedicated method parse_be_length which has less overhead.

Improved static_left_pad to not always zero init an array, and added the special case for data.len() == N to avoid a saturating sub.

Helps with #4956

[github-actions]

Lines of code report

Total lines added: 77
Total lines removed: 0
Total lines changed: 77

Detailed view

+------------------------------------+-------+------+
| File                               | Lines | Diff |
+------------------------------------+-------+------+
| ethrex/crates/common/rlp/decode.rs | 730   | +77  |
+------------------------------------+-------+------+

[github-actions]

Benchmark Block Execution Results Comparison Against Main

Command	Mean [s]	Min [s]	Max [s]	Relative
base	61.106 ± 0.182	60.893	61.440	1.01 ± 0.00
head	60.737 ± 0.210	60.463	60.981	1.00