vscode: add configurable server context limit

[hoyyeva]

Summary

Add a configurable maximum context length for Ollama servers.

Users can set Ollama: Max Context Length to the same value as the server's OLLAMA_CONTEXT_LENGTH. The extension caps the model context reported to VS Code without exceeding the model's native context window.

This replaces the need to patch the installed extension with a hard-coded limit.

The change also:

• Supports global and per-provider context limits.
• Refreshes the model list when the setting changes.
• Logs the configured cap in the Ollama output channel.
• Documents setup in the README.
• Preserves existing behavior when the setting is not configured.

Oversized one-shot requests that bypass VS Code's advertised limit remain unchanged and are tracked separately in #33.

Fixes #31.

Testing

• Added tests for discovered, configured, fallback, and very small context windows.
• Added tests for provider-over-global precedence.
• Added tests for invalid context values.
• Verified manually with OLLAMA_CONTEXT_LENGTH=8000.
• Confirmed the output channel reports the configured value:

capped at 8000 context tokens

• All 25 tests pass:

npm test