solidity: use shifts instead of multiply in LEB128 length decoder

[deuszx]

Summary

* 128 and * val on uint256 each cost 5 gas where << 7 costs 3. The Yul optimizer probably already folds * 128 to << 7, but rewriting the inner loop in terms of shift/OR makes the intent explicit and removes the question.

Test Plan

Functionally equivalent: existing multi-byte length round-trip coverage (vec i64 of length 140) exercises this path.

[deuszx]

Under 0.8 checked arithmetic, the old power *= 128 / result += … would revert on a malformed length field with enough continuation bytes to overflow uint256 (~37+ bytes). The new code can't revert: once shift >= 256, val << shift is 0 in Solidity and result |= 0 is a no-op. So malformed input that previously reverted now silently decodes to a truncated length instead.