Correct encoding examples, padding lengths and decoder alphabet documentation

[OskarEichler]

Summary

Correct the Base64 documentation's encoded values, newline lengths, padding explanations, and accepted alphabets. Regenerate the corresponding RBS documentation with the existing rbs:annotate task. No runtime implementation or type signatures change.

Reproduction

Current documentation says Base64.strict_encode64('123') returns "MDEy", but it returns "MTIz". The same shifted input affects twelve examples. Base64.encode64("\x00").size is 5, not 4, because the newline counts toward the length.

The unpadded-output sections incorrectly describe trailing padding, and the decoder overview incorrectly says every decoder accepts missing padding. / belongs to the standard alphabet. URL-safe decoding also accepts the standard alphabet (and mixed alphabets), so testing a single + or / demonstrates an invalid length rather than an alphabet restriction.

Verification

Using Ruby 4.0.6 through rbenv:

• Existing rake test: 9 tests, 73 assertions, no failures/errors.
• Existing rake rbs:test: 6 tests, 35 assertions, no failures/errors.
• External, uncommitted verification: 2,636 checks, including 70 documented examples and frozen-input round trips for input lengths 0–512.
• Runtime token comparison against the base revision: unchanged.
• rake rbs:annotate, Ruby syntax check, gem build, and git diff --check pass. Gem build retains the existing metadata warnings.

No tests were added or modified, per the requesting project's contribution constraint. Verification was local on macOS/Ruby 4.0.6; the full supported Ruby/OS matrix was not run locally. Existing documentation fixes already on main, including #46's signature comments, are not duplicated.

Breaking changes

None. This changes documentation only; accepted input, output, signatures, dependencies, and version are unchanged.