Unify AI provider selection across all features

Every LLM-backed feature (Post-Processing, Proofreading, Title
Generation, Property Filler) now reads from one CHAT_PROVIDERS
registry instead of 5 hand-duplicated provider lists, so they always
offer the exact same providers in the exact same order.

- Add Groq as a 5th chat-completion provider (was transcription-only
  despite its OpenAI-compatible API and an already-collected key).
- Provider dropdowns disable options with no API key configured
  (shown as e.g. "Anthropic (no API key)") instead of silently
  failing at runtime when picked.
- Collapse 3 byte-for-byte duplicated resolveProvider() methods
  (Proofreader, TitleGenerator, PropertyFiller) plus AudioHandler's
  getPostProcessingApiKey() into one resolveChatProvider() helper
  (src/ProviderResolver.ts).
- Fix Custom provider: previously Proofreading/Title/Property silently
  reused Post-Processing's postProcessingUrl with no visible field to
  set it. Renamed to a shared customApiUrl field with its own UI
  (createCustomProviderFields), now exposed on every tab that offers
  Custom, plus rows on the API Keys tab. Migrates old data.json values
  automatically.
- Update README settings reference and getting-started sections to
  match.

No behavior change for existing single-provider setups; migration
handles the postProcessingUrl -> customApiUrl rename transparently.

src/AIService.ts

@@ -5,7 +5,7 @@ export interface ChatCompletionOptions {
 	apiKey: string;
 	model: string;
 	endpoint?: string;
-	provider?: "anthropic" | "openai" | "openrouter" | "custom";
+	provider?: "anthropic" | "openai" | "openrouter" | "groq" | "custom";
 	temperature?: number;
 	maxTokens?: number;
 }