Fix Portuguese coverage gaps: synonyms, translations, frequency, gerunds

- Extract sense-level synonyms/antonyms from Wiktionary (not just top-level),
  dramatically improving Portuguese synonym coverage
- Add bidirectional translation lookup so pt→en queries find en→pt entries
  stored by English Wiktionary loader, with reverse index for performance
- Fix frequency loader: auto-detect column order (count\tword vs word\tcount),
  auto-detect ISO-8859-1 encoding and convert to UTF-8, insert missing words
  instead of only updating existing ones
- Switch WordNet download from 3.1 to 3.0 to match WOLF/OMW synset offsets
- Add OpenSubtitles-derived Portuguese frequency list (hermitdave/FrequencyWords)
- Increase affix expansion cap from 30→80 forms per word for Portuguese gerunds

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

internal/loader/affix.go

@@ -205,9 +205,9 @@ func parseDicFile(path string) ([]dicEntry, error) {
 }
 
 // maxFormsPerWord caps the number of inflected forms generated per base word.
-// Portuguese and French .aff files can produce hundreds of cross-product forms
-// per word; most are valid but the combinatorial explosion bloats the DB.
-const maxFormsPerWord = 30
+// Portuguese verbs can have 60+ valid conjugations (gerunds, subjunctives, etc.),
+// so the cap must be generous enough to avoid cutting off common forms.
+const maxFormsPerWord = 80
 
 func expandWord(word, flags string, rules map[string]*affixRuleSet) []string {
 	seen := map[string]bool{word: true} // base form already in DB