fix: guard address validation against non-string values

[ayushsingh82]

Summary

• is_valid_address() calls re.match() directly on the address field. If address in data.json isn't a string (e.g. a JSON number, null, or a list), this raises an uncaught TypeError.
• validate_token_directory() doesn't catch exceptions from validate_token_data(), so this TypeError propagates all the way up to main()'s top-level except Exception, printing Unexpected error: ... and aborting validation for every token in the run — not just the malformed one.
• Every other field validated in validate_token_data() (name, symbol, chainId, decimals) already checks isinstance() before further validation. The top-level address field and crossChainAddresses[chain_id].address did not follow this same pattern.
• Adds the missing isinstance(address, str) guard in both places, matching the existing style, so a bad address field now produces a normal, scoped "Invalid address: ..." error for that token instead of crashing the whole script.

Test plan

• Reproduced the crash in isolation: validate_token_data({"address": 12345, ...}, ...) raised TypeError: expected string or bytes-like object before the fix.
• After the fix, the same input returns a normal error list including "Invalid address: 12345" — no crash.
• Same fix verified for crossChainAddresses[chain_id].address (e.g. {"1": {"address": 999999}}).
• Ran the address-format check against all 89 existing mainnet/*/data.json files — identical results before and after (no regression on valid string addresses).
• uv run ruff check and uv run ruff format --check pass on the changed file.