Pre-existing files that needed reformatting to satisfy the CI black check
after upgrading to black 26.1.0.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Fixes #395. All workflow tools share a single singleton instance in
server.py. When multiple agents call the same tool concurrently, they
race on shared self.* mutable state at asyncio await yield points.

Changes:
- workflow_mixin.py: add asyncio.Lock per tool instance; split
  execute_workflow into _execute_workflow_locked so subclasses can
  call the locked body without re-acquiring the lock
- consensus.py: add _sessions dict keyed by continuation_id to isolate
  original_proposal, models_to_consult, accumulated_responses per caller;
  restore work_history/consolidated_findings from conversation memory on
  continuation steps (base class logic that the custom consensus loop
  bypassed); fix off-by-one in customize_workflow_response status
  condition (step_number < total_steps, not < total_steps - 1)
- all other workflow tools (analyze, codereview, planner, precommit,
  refactor, secaudit, testgen, tracer): update store_initial_issue call
  signature to match updated base class

Allow the MCP server to run over HTTP using SSE transport in addition
to the default stdio. Controlled via MCP_TRANSPORT, MCP_HOST, and
MCP_PORT environment variables.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add gemini-3.1-pro-preview (released 2025-02-19) with intelligence
score 19, 1M context, 65K output, extended thinking, and full
capability support. The "pro" alias now points to 3.1.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

SimpleTool subclasses (chat, clink, apilookup, challenge) are
singletons that store per-call state on self during execute(). Under
concurrent SSE connections, two sessions could clobber each other's
_current_arguments, _model_context, and _current_model_name.

Add _concurrency_lock (asyncio.Lock) matching the pattern already
used by WorkflowMixin, serialising concurrent calls to the same tool.

Also add a port-availability check in _run_sse() so the server fails
fast with a clear error if a stale process holds the SSE port.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>