Boost exact-character matches over ASCII-folded matches

[henrik242]

Summary

• Adds a non-folded raw sub-field on collector.name that indexes names with only lowercasing (no ASCII folding, no German normalization)
• Adds boosted should clauses in both short and full search queries so documents whose unfolded name matches the query score higher
• Fixes ranking where e.g. searching "Lærdal" would return "Lardal" with equal or higher prominence, because both collapsed to the same token after ASCII folding

Behavior

• "Lærdal" → "Lærdal" ranks above "Lardal" (raw field match on the query)
• "Lardal" → "Lardal" ranks above "Lærdal" (raw field match on the query)
• "lardal" → still finds both via the folded fields (broad recall preserved)

No existing behaviour changes. german_normalization is left in place. The raw field only adds score - it never removes matches.

Test plan

• New test testExactDiacriticsOverFolded validates ranking directly
• All existing tests pass (./gradlew test)

[lonvia]

This only solves the issue on the full names, not during search-as-you-type (try searching for Lård). You'd need yet another secondary index for the prefix matchers. That's quite expensive and the reason I rejected this approach.

In general, a second index is only required, when the expected results don't show up at all in the result list. If it is a more a matter of in which order the results are shown, then reranking results is the cheaper approach.