fix(decode): reject address words with non-zero upper bytes

[devorun]

Summary

decode_address returned the last 20 bytes of the ABI word and silently discarded the upper 12, so a malformed word such as 0xffff…ffff<20-byte-addr> was accepted as a valid address:

return to_checksum_address(decode_word(result, name)[-20:])

This is inconsistent with the rest of _decode.py, which deliberately rejects non-canonical words — decode_bool rejects values other than 0/1, and decode_u64 rejects overflow. A canonical ABI address word left-pads the 20-byte address with 12 zero bytes, so non-zero upper bytes signal a malformed response and shouldn't be quietly dropped.

Fix

Reject a non-zero upper 12 bytes with the same ValueError style the sibling decoders use:

word = decode_word(result, name)
if word[:12] != b"\x00" * 12:
    raise ValueError(f"{name} result has non-zero upper bytes for a 20-byte address")
return to_checksum_address(word[-20:])

Canonical words decode exactly as before.

Tests

Adds tests/test_decode.py (these internal decoders had no dedicated test file): the address strictness fix — canonical word accepted, non-zero upper bytes rejected, wrong length rejected — plus sibling guards for decode_bool / decode_u64 / decode_word.

pytest passes (272 passed, 36 skipped; +6). ruff check and ruff format --check are clean.