feat: fuzzy search with relevance ranking and highlight

[hermes-alby]

Summary

Replace the simple substring-only search with a proper fuzzy search that scores and ranks results by relevance.

Before

• Only matched exact substring in title + description
• No scoring or ranking — results appeared in original order
• Typing "nostr wallet" would return nothing unless that exact phrase existed

After

• Multi-level scoring: exact match > startsWith > word boundary > substring > fuzzy subsequence
• Multi-word queries — each word is matched independently, so "nostr wallet" finds apps mentioning both
• Relevance ranking — title matches are scored 100x higher than description matches, so results appear in order of relevance
• Match highlighting — matching text is highlighted with a yellow background in search results
• Zero dependencies added (~150 lines of pure TypeScript)

How scoring works

• 100 = exact match
• 90 = starts with query
• 80 = word boundary match
• 70 = mid-string substring match
• 30-75 = multi-word fuzzy match
• 40-59 = single-word fuzzy subsequence