feat(providers): split microsoft api_key_env from github-copilot (#83)

Both `microsoft` (GitHub Models / Azure AI Inference at
models.inference.ai.azure.com) and `github-copilot` (the IDE subscription
product) previously declared `api_key_env = "GITHUB_TOKEN"`, so a single
PAT silently activated both providers — users with only the IDE
subscription saw GitHub Models entries appear in their model picker
without intent, and vice versa.

Rename `microsoft` to `GITHUB_MODELS_TOKEN`. `github-copilot` keeps
`GITHUB_TOKEN` as the established convention for the IDE side.

Existing users who set `GITHUB_TOKEN` to use the GitHub Models endpoint
will see the `microsoft` provider become unavailable until they set the
new env var. The librefang daemon will be updated separately to
recognize `GITHUB_TOKEN` as a deprecated fallback for `microsoft` during
a migration window, mirroring the legacy-fallback infrastructure added
in librefang/librefang#3279.

Refs librefang/librefang#3278

providers/microsoft.toml

@@ -1,9 +1,15 @@
 # microsoft — auto-generated from OpenRouter API
+#
+# This entry is GitHub Models / Azure AI Inference (https://models.inference.ai.azure.com),
+# distinct from the `github-copilot` IDE-subscription product even though both
+# physically accept a GitHub Personal Access Token. Each now has its own env so
+# users with only one product configured don't see the other appear in the model
+# picker. See librefang/librefang#3278.
 
 [provider]
 id = "microsoft"
 display_name = "Microsoft"
-api_key_env = "GITHUB_TOKEN"
+api_key_env = "GITHUB_MODELS_TOKEN"
 base_url = "https://models.inference.ai.azure.com"
 key_required = true