This fixes two correctness bugs in the crypto_addresses validators. Both are covered by new regression tests, and the full test suite still passes.

1. btc_address rejects valid testnet segwit addresses

The segwit branch is selected when the value starts with bc or tb, but the regexp only matched bc or tc:

re.compile(r"^(bc|tc)[0-3][02-9ac-hj-np-z]{14,74}$").match(value)
if value[:2] in ("bc", "tb")
else _validate_old_btc_address(value)

Because the gate accepts tb while the pattern expects tc, every valid testnet bech32 address is rejected:

btc_address("tb1qw508d6qejxtdg4y5r3zarvary0c5xw7kxpjzsx")  # ValidationError, should be valid

tc is not a valid Bitcoin HRP and is unreachable through the value[:2] in ("bc", "tb") gate, so it is a typo for tb. Fix: (bc|tc) → (bc|tb). Mainnet bc1... is unaffected.

2. eth_address accepts non-hex "checksum" addresses

_validate_eth_checksum_address checked only the length of the stripped address, not that its characters were hexadecimal. The EIP-55 loop only constrains the case of letters, so caseless / non-hex characters pass it vacuously:

eth_address("0x" + "*" * 40)  # True, should be invalid

Fix: require the stripped address to be exactly 40 hex digits before running the checksum (this also subsumes the previous length check).

Tests

• test_btc_address: added two valid testnet bech32 addresses (BIP173 vectors) to the valid fixtures.
• test_eth_address: added a non-hex input to the invalid fixtures.

Both new cases fail on master and pass with this change. ruff reports no issues and the full suite (898 passed) is green.