Improve counterexamples diagnostics

[ydah]

This PR closes several gaps between Lrama's counterexamples diagnostics and Bison's behavior.

Summary

• Add warning-category support for -Wcounterexamples / -Wcex, and honor -Wnone / --warnings=none.
• Emit a follow-up note suggesting -Wcounterexamples when conflicts exist but counterexample warnings are not enabled.
• Render counterexamples in a Bison-style structure:
  • Example: when both sides share the same witness
  • First example: / Second example: when the witnesses differ
  • derivation sections for each side
• Compute counterexamples per conflict token instead of only using the first token in a conflict.
• Avoid misleadingly merging multi-token counterexamples when the witness differs by lookahead.
• Render counterexamples in situ in the state/automaton report.

Refs:

• https://www.gnu.org/software/bison/manual/html_node/Counterexamples.html
• https://www.gnu.org/software/bison/manual/html_node/Understanding.html