Enrich the lexer to support the entire CEL lexicon

[mcy]

This PR allows our lexer to parse the following CEL productions:

• Suffixed numbers, such as 42u.
• Prefixed strings, such as r"", b"", and rb"".
• Triple-quoted strings, such as """foo""".
• < and > are separate tokens (which the parser now fuses in some cases).
• in is now a keyword.
• Added all CEL punctuation, such as arithmetic operators.

These all generate diagnostics in the legalizer. I have also moved things out of the lexer and into the legalizer to simplify verification of constraints.

This makes all of the changes necessary so that we can support full CEL expressions anywhere in Protobuf in the future. Doing this now is ideal, because making this work (especially with <...> matching) will only get harder as time goes on.

This change also causes us to reject some pathological Protobuf code, involving implicitly-concatenated strings. For example, option foo = """"; is valid Protobuf, but it looks like an incomplete triple-quoted string. Rejecting this is almost certainly fine, because this very likely does not exist in the open source corpus.