Add missing -ize/-ise words and fix two reversed word-list entries

[niksedk]

Fixes the miss reported in SubtitleEdit/subtitleedit#13583 ("fantasize" => "fantasise") — and the rest of the same gap.

Word list

The shared list only covered 66 -ize verbs, so maximize, patronize, authorize, characterize, optimize, socialize, synchronize and ~95 more were missed as well. This adds 608 pairs:

• 100 regular -ize/-ise verbs with all conjugations (-ize/-ized/-izes/-izing)
• the conjugations missing for sanitize and deodorize (only the -ed form was listed)
• agent nouns: synthesizer, sterilizer, vaporizer, digitizer, atomizer, nebulizer, sanitizer, deodorizer
• -ization/-isation nouns, including for verbs that were already listed (specialization, utilization, mobilization, hospitalization, …)

The list drives both directions, so every pair has to be a genuine both-ways variant. Left out on purpose:

• words that are -ise in both varieties — advertise, improvise, exercise, surprise, compromise, televise, supervise, capsize, prize, … — adding those would make British to American emit advertize
• -yze/-yse verbs (catalyze), because catalyses/analyses collide with the noun plurals
• glamorize (the -our stem makes the British form ambiguous), plus rare/one-sided ones like winterize, notarize, exorcize, systemize

Two bugs found while validating

• us="dustbin" br="garbage can" was reversed — American to British turned dustbin into garbage can, and British to American did the opposite. Now us="garbage can" br="dustbin" (+ plural).
• jewelery preceded jewelry, and for BR→US the first rule matching jewellery wins, so jewellery became the misspelling jewelery. Reordered, with a comment on why the order matters.

Keeping it fast

BuildProposals() runs on the UI thread in the window constructor, and 50% more rules would have made that worse. Each rule pattern starts with \b, so its first word can only ever match a whole word of the input — a line that does not contain that word can never match the pattern. Rules are now indexed by that first word, and a line runs only the handful of rules whose words it actually contains instead of all 5541. RegexOptions.Compiled is dropped along with it, since almost no rule runs per line now.

Verified byte-identical to the old full scan over a 49,560-line corpus (every word alone / in a sentence / uppercased / titlecased / glued to punctuation, tags and digits, plus 20,000 random multi-word lines) in both directions:

	before	after
construction (one direction)	96 ms	12 ms
convert 2000 lines	770 ms	1.5 ms

Checks

• both plugins build clean
• fantasize/fantasized/fantasizes/fantasizing + FANTASIZE/Fantasized convert in both directions
• advertise, improvise, capsize, prize, surprise, exercise, compromise still untouched
• <font color="#ff0000"> attribute still not Britishized

Needs a release dispatch for AmericanToBritish and BritishToAmerican (both embed the shared list), then a se5-plugins.json bump.

🤖 Generated with Claude Code