feat(config): make API base URLs and model names configurable via env

[subashrijal5]

Summary

Makes the Anthropic and OpenAI API base URLs configurable via environment
variables (with sensible defaults), and reworks model-name resolution so each
provider has its own configurable model name. This unblocks using proxies,
gateways, or self-hosted/OpenAI-compatible endpoints without code changes.

Changes

• Configurable base URLs — add anthropic_base_url and openai_base_url
  to AppConfig, sourced from ANTHROPIC_BASE_URL / OPENAI_BASE_URL and
  defaulting to the official endpoints (https://api.anthropic.com,
  https://api.openai.com/v1).
• Per-provider model names — replace the gemini/openai-only model selector
  with a _default_model_name() resolver driven by AI_PROVIDER, reading
  GEMINI_MODEL / OPENAI_MODEL / ANTHROPIC_MODEL. Fixes a gap where the
  claude provider had no model env var and silently fell back to the Gemini
  default.
• Relaxed validation — drop the hard-coded OpenAI model whitelist (it only
  permitted gpt-4.1-nano, which conflicts with custom/gateway endpoints).
  OpenAI and Claude now just require a non-empty model name; the Gemini
  whitelist is unchanged.
• Docs — document the new variables in .env.example.

Defaults / backwards compatibility

All new variables are optional and fall back to the previous behavior, so
existing Gemini-based setups are unaffected.

Notes

These values are now available on get_config(), but no Anthropic/OpenAI
client is wired into the generation pipeline yet — the active path remains
Gemini. This PR is the config groundwork for that future wiring.