feat: add Streamable HTTP MCP and keyless Parallel search example

[georgeatparallel]

Notes

Adds Streamable HTTP to rg.mcp, so Rigging can connect directly to hosted MCP servers such as Parallel Search MCP. The new examples/parallel_search.py searches the web without a model or Parallel API key:

python examples/parallel_search.py "dreadnode Rigging MCP documentation"

The tools guide also shows how to attach the discovered search and fetch tools to a chat pipeline. Anonymous Parallel access has lower rate limits; model inference still requires the configured generator's credentials.

Key Changes:

• Load Streamable HTTP tools through the existing MCP client and Rigging tool API.
• Forward headers and HTTP/streaming timeouts to the MCP SDK.

Added:

• A keyless Parallel search example and usage documentation.
• Transport tests for discovery, tool execution, timeout forwarding, cleanup, cancellation, and existing transport selection.

Changed:

• Raise the MCP SDK minimum to 1.10.0, which supports Streamable HTTP; keep the locked SDK version at 1.21.1.
• Update the generated tools API documentation and README example list.

Removed: None.

Validation:

• Full pytest suite, Ruff, and mypy.
• Transport tests and live anonymous search/fetch with the minimum MCP SDK 1.10.0 and locked SDK 1.21.1.
• Executed search and fetch through rg.mcp discovery and Tool.handle_tool_call, receiving source URLs and excerpts. Observed rigging/3.3.5 User-Agent headers and no Authorization header on the requests.
• Ran the standalone search example with isolated credentials.

Btw - I'm a devrel engineer at Parallel