v0.14.0

Added

• DRFMCPToolset(max_retries=...) — each tool's retry budget: how many times a
  ModelRetry (malformed arguments, a service-raised validation error) is fed
  back to the model before the run aborts. Defaults to 1, matching
  pydantic-ai's own function-tool default.

Changed (breaking)

• DrfMcpToolset is renamed DRFMCPToolset, matching the capitalized
  acronyms of its sibling classes (MCPServer, AGUIServer) and PEP 8's
  CapWords convention. The class is built internally by the view from the
  DRF_MCP_SERVER setting, so only code importing it directly from
  django_ag_ui.integrations.drf_mcp needs the one-line rename; no alias is
  kept.

Changed

• DRFMCPToolset now subclasses pydantic_ai.toolsets.AbstractToolset
  directly (the documented extension point) instead of ExternalToolset,
  building its tool definitions kind="function" from the start. Previously
  it inherited from a base class that models the opposite of in-process
  execution (external tools are deferred to the client) and re-stamped every
  tool definition back to kind="function" per run — the version-fragile seam
  behind the historically tight <2 pydantic-ai pin. Public API and tool
  behaviour are otherwise unchanged.

Fixed

• A bridged tool's ModelRetry (malformed arguments, a service-raised
  validation error) now actually reaches the model to self-correct, as
  documented. ExternalToolset pinned every tool's retry budget to 0, so in
  a real agent run the first ModelRetry aborted the run with
  UnexpectedModelBehavior instead of retrying. Pinned by a full agent-run
  integration test.