Preprocess to remove brackets and other punctuation before attempting grounding

[caufieldjh]

A recurring issue: instances where the LLM adds extra formatting/punctuation tend to break things. Example:

  patient_weight:
    value: <45 kg; 35 kg>
    unit: AUTO:%3Ckg%3B%20kg%3E
    percentile: <46%; 6%>
    zscore: <-0.10; -1.55>

This will probably need to be handled in the SPIRES implementation itself as a post-processing step.

Originally posted by @caufieldjh in #319 (comment)