refactor ai video gen

[P3il4]

refactored ai video generation into a single service

• created main service file similar to image service
• added two guards against bad duration and dimension values (from wip: ai video gen ai refactor #2698 )
• changed costs logic from pulling costmaps to having it in models.ts and using costOverrides
• added duration and seconds values to /models/details of openai models for consistency

smaller stuff:

• added sora pro xxl ($0.70 tier) https://developers.openai.com/api/docs/models/sora-2-pro
• added options.test_mode for video missing from docs
• added exact model id match because of openai/together sora alias

tested with a mock server implementation of openai video and togetherai video by changing baseurl
.getMonthlyUsage() logged costkeys remain the same
legacy driver hints openai/together-video-generation still work
possibly breaking change for apps: in togetherai durationSeconds is now an array instead of a number
and some togetherai models ids are now lowercase
togetherai:ByteDance/Seedance-1.0-pro -> togetherai:bytedance/seedance-1.0-pro

[ProgrammerIn-wonderland]

GitHub quite literally will not let me merge this PR because it timed out the first time

[ProgrammerIn-wonderland]

you might need to just close this one and make another one if it doesn't resolve itself

[ProgrammerIn-wonderland]

but everything seems fine