Improve approximate string search

[carlgieringer]

Sometimes our approximate string search library can't find a more complete match, even though it seems like it should be possible with fewer errors.

E.g. for the quotation

Lex Fridman

(00:21:33) And you think that kind of empathy that you referred to, that requires moral courage?

(from https://lexfridman.com/robert-f-kennedy-jr-transcript/)

https://github.com/robertknight/approx-string-match-js#readme returns {start: 19933, end: 20035, errors: 21}, which corresponds to "33) And you think that kind of empathy that you referred to, that requires moral courage?' when it should be possible for this quotation to match the text with fewer errors:

                Lex Fridman
                (00:21:33)
                And you think that kind of empathy that you referred to, that requires moral courage?

At the least it could match more of the timestamp and have an equal number of edits with a larger match.

The problem may be the leading whitespace. We could probably remove this for approx-string-match-js, but we'd need to figure out how to accommodate that in dom-anchor-text-position toRange.

[carlgieringer]

Possibly we could provide our own toRange that provides a custom dom-seek approach that will:

• Remove leading whitespace
• Collapse more than two newlines into two newlines.

https://github.com/tilgovi/dom-anchor-text-position/blob/6502c48aff7f3f0ce3bb225d1c04ca8624c3b88f/src/index.js#L34-L58