Base64 decoder

Decodes Base64 back to text, accepting the URL-safe alphabet and restoring padding that has been stripped — the two things that make a decoder reject perfectly valid input.

Padding is the usual reason a decoder refuses input it should accept. The `=` characters exist to bring the length to a multiple of four, and many systems strip them because they are redundant — JWTs always do. A decoder insisting on them fails on good data. This restores them and rejects only a length that is genuinely impossible. The other common surprise is decoding something that was never text: Base64 encodes bytes, so decoding an encoded image gives binary rubbish rather than an error, because there is nothing invalid about it.