evmnfts, evmtokens: utf-8-clean strings from chain

[pro-wh]

values that come from chain: use our SanitizeString, as they may contain postgresql-displeasing nul characters

edit: as well as values that come from servers 🙄

[mitjat]

Thank you!

Worth mentioning why we're implementing this PR:
We found invalid metadata JSON in the wild (in particular, it was invalid UTF-8 in that it contained \x00, but the Go JSON parser didn't balk at that, but Postgres does).
Hope I got that right :)

[mitjat]

:(
It does seem low risk.
We could have another wrapper, ensureValidJSON(), that returns the input as-is if it's valid JSON, otherwise logs a warning and returns {} or null. Your call on whether that's warranted. I think you already made the call with this comment :)

[pro-wh]

ok turns out the round trip through json unmarshal+marshal is enough to filter out bad UTF-8 and NUL bytes. updated the comment. I'll keep the SanitizeStringP here out of principle

[pro-wh]

e2e regression flaked, http 500 from ipfs gateway

[pro-wh]

oh clarification about @mitjat 's explanation: \x00 NUL bytes are valid UTF-8, but they're not allowed in PostgreSQL due to a technical limitation.

incidentally, they're also not allowed in JSON, as strings aren't allowed to contain control characters (\uxxxx escapes of them are allowed), and the grammar doesn't allow them anywhere else either.