Centralize provider credentials and add model discovery

[joshtech90]

Summary

• store Groq/OpenAI API keys once in a central Providers tab
• let transcription and rephrasing independently choose provider and model
• load the provider model catalog from the OpenAI-compatible /models endpoint without blocking the UI
• feature-filter and cache model lists, with maintained offline fallbacks
• preserve separate credentials only for genuinely separate custom OpenAI-compatible endpoints
• migrate legacy duplicated URL/key fields to schema v2 and remove them after migration

Migration behavior

• existing known-provider keys are deduplicated
• old display suffixes such as (groq) are cleaned from model IDs
• an untouched secondary OpenAI default automatically reuses an already configured transcription provider
• hotkeys, prompts, transformations and unrelated settings remain unchanged

Verification

• ruff check app tests run.py
• mypy app/core app/services app/audio app/hotkeys app/ui
• 136 pytest tests
• full Qt runtime smoke: startup, warnings, transcription, LivePrompt, failure fallback and batch transcription
• live Groq catalog check: 2 transcription and 8 rephrasing models detected
• visual review of Providers, Transcription and Rephrasing tabs on macOS