@erikcorry recently remarked that the maximal_binary_length_from_base64 could be confusing. This PR attempts to better document it and justify how it works.

The contract is thus:

1. Constant time execution (very fast).
2. Exact result if there is no ignorable characters (spaces, linefeed...).
3. Exact result even if there are padding characters

The reason why it is reasonable is that most base64 inputs have very few ignorable characters. And if you add lots and lots of spaces to your base64 payloads, you are probably not focused on minimizing memory usage.