Fix various numeric literal token errors

[c42f]

A fairly big refactor of numeric literal tokenization error cases and a couple of other tokenizer errors ported from the flisp code.

• For hexfloat, emit a more specific errors when the p suffix is missing.
• For octal, hex and binary, add errors for trailing invalid digits or identifier characters like 0b123 and 0xenomorph
• Emit an error for ambiguous numeric constants with dot suffix vs juxtuposition like 1.(
• Emit an error for underscore directly after dot as in 1._
• Emit an error for hexfloat without digits 0x.p0
• Add an invalid operator error for <--- to follow compatibility with the reference parser.

Refactor and regroup all the numeric tokenization tests which were confusingly grouped.

Fix #57 #172