fix(l1,l2): make RLP integer impl_encode single-pass and zero-allocation

[Forostovec]

Reworked impl_encode in crates/common/rlp/encode.rs to compute the first non-zero byte in a single pass and to encode using direct slices of the original big-endian array. This eliminates the previous double scan and avoids allocating/copying into ArrayVec, which was unnecessary for encoding the integer payload. As a direct consequence, this change also fixes a correctness bug where u128 encoding could panic at runtime due to relying on ArrayVec<[u8; 8]> and attempting to extend it with 16 bytes. The new logic preserves RLP semantics for zero (0x80), single-byte values in the 0x01..=0x7f range (encoded as-is), and multi-byte integers (prefixed with 0x80 + length followed by the trimmed big-endian bytes), aligning with how other parts of the module handle length-prefixed payloads without redundant work.

[edg-l]

There is already a pr open which improves the encoding #5350

[Forostovec]

There is already a pr open which improves the encoding #5350

Ops, then i will close this