feat: add Gemini Flash-Lite as a selectable AI subtitle-translation model

[mvanhorn]

Summary

Gemini Flash-Lite is now a selectable model for AI subtitle translation. The Gemini request path is parameterized by model id instead of baking gemini-2.5-flash into the base-URL constant, and the new GEMINI_FLASH_LITE enum value routes through the same translateGemini implementation with gemini-2.5-flash-lite.

Why this matters

#251 asked for Flash-Lite as an option for faster, cheaper subtitle translation; @silentbil acknowledged it on 2026-05-28 and the reporter pinged again on 2026-06-11 with no PR in sight. This covers the model-addition half of the issue; the advanced generation settings (temperature, top-p, thinking budget) are a separate piece and stay open. Persistence needs no changes since the enum is stored by name with a safe fallback, so existing users see no migration.

Testing

Unit test covers the enum round-trip and the model-id mapping; the settings screen's model label, disclaimer text, and model dialog all render the new option through the extended exhaustive when blocks.

Refs #251

[silentbil]

Hi @mvanhorn
This request was for gemini 3.1 flash lite
while this PR just added a 2.5 flash lite.
I compared 2.5 vs 2.5 lite in the past. and 2.5 gave better quality at almost same price.
i wanted to check out 3.1 flash lite which could make things better, maybe change your pr to use it