ENH: Avoid repeated name normalization during matching

[GibranAlfa]

Resolves #460

Normalize each source and target name once per matching request, then use RapidFuzz extraction to calculate scores and select the top three matches.

Time to compare 1550 wells is reduced from 159 seconds to 220 miliseconds 🤯.

Checklist

• Tests added (if not, comment why)
• Test coverage equal or up from main (run pytest with --cov=src/ --cov-report term-missing)
• If not squash merging, every commit passes tests
• Appropriate commit prefix and precise commit message used
• All debug prints and unnecessary comments removed
• Docstrings are correct and updated
• Documentation is updated, if necessary
• Latest main rebased/merged into branch
• Added comments on this PR where appropriate to help reviewers
• Moved issue status on project board
• Checked the boxes in this checklist ✅

[tnatt]

Nice speedup 🥳
I just have some comments regarding tests - I feel like we might not test too many variants of different strings. Probably worth adding some more.