UDF Cleanup

Writing UDFs needs to be more robust in face of user errors.

• Don't crash
• Report all user errors (both parse and bind) so that they're actionable.

Things to cleanup in UDFs:

1. We immediately crash on unrecognized types in the signature (whenever GetFromStringOrNull returns null)
2. We double parse the function body (UDFDefinition has a string body, it should store the TexlNode or even ParseResult). We're going from TexlNode back to String via TexlPretty ... and that injects "<error>" into the string contents on parse errors, which then get re-parsed! This also means that error spans don't align up with the user's original input text.
3. CheckWrapper - don't do deferred binding. Bind immediately. This may require 2 passes for recursion (first pass to populate signatures, second pass to handle bodies)
4. Better error handling - ensure we return all errors

Trace UDF processing through UDFDefinition, CheckWrapper, GetFromStringOrNull, TexlNode, ParseResult, and TexlPretty. Start by mapping where parsing and binding occur, then verify that unrecognized types and parse or bind failures are reported without crashes, preserving the user's original text and error spans.