fix(bigint): canonicalize from_octets with leading zero limbs

[hyfdev]

Closes #3809.

Problem

BigInt::from_octets only removed one zero most-significant limb. Inputs with more than one zero limb retained an invalid logical length, so non-JS backends could compare equal values incorrectly. An all-zero multi-limb input could also form a negative zero and make to_string panic.

Fix

Use the existing normalize_len helper to compute the canonical logical length after constructing the limbs.

Tests

Extend the existing whitebox from_octets coverage with the issue input and multi-limb zero inputs for both default and negative signums. Each case checks the internal invariant plus string, equality, and comparison behavior.

Validation

• moon check --deny-warn --target all
• moon test --target all
• moon test --release --target js,wasm,wasm-gc
• moon test --target native --release
• JS builtin string tests on wasm-gc in debug and release
• moon info --target wasm,wasm-gc,js,native (no interface diff)
• moon fmt --check
• moon bundle --all