fix(sdk): use ephemeral ports for Parlant servers

[darvell-thenvoi]

What changed

Parlant examples and test fixtures now start in-process Parlant servers with OS-assigned ephemeral ports:

p.Server(
    port=0,
    tool_service_port=0,
    nlp_service=p.NLPServices.openai,
)

This avoids collisions when multiple Parlant agents run at the same time, such as the Tom and Jerry examples.

The generic example runner also now exposes Parlant directly:

• uv run python examples/run_agent.py --example parlant
• uv run python examples/run_agent.py --example parlant --streaming

The Parlant docs and runner error message now point repo users at the current uv sync --extra dev setup.

Why

Parlant's default server ports are fixed. Starting more than one in-process Parlant server can make the second process fail to bind. Passing port=0 and tool_service_port=0 lets the OS pick available ports for each server, so the examples and E2E setup do not need hardcoded or manually coordinated port numbers.

Validation

uv run ruff check \
  examples/parlant/*.py examples/run_agent.py \
  src/band/adapters/parlant.py src/band/integrations/parlant/tools.py \
  tests/adapters/test_parlant_adapter.py tests/e2e/adapters/test_parlant.py \
  tests/framework_conformance/test_tool_name_drift.py \
  tests/integrations/parlant/test_tools.py tests/test_run_agent.py

uv run ruff format --check \
  examples/parlant/*.py examples/run_agent.py \
  src/band/adapters/parlant.py src/band/integrations/parlant/tools.py \
  tests/adapters/test_parlant_adapter.py tests/e2e/adapters/test_parlant.py \
  tests/framework_conformance/test_tool_name_drift.py \
  tests/integrations/parlant/test_tools.py tests/test_run_agent.py

uv run --extra dev pytest \
  tests/test_run_agent.py \
  tests/adapters/test_parlant_adapter.py \
  tests/integrations/parlant/test_tools.py \
  tests/framework_conformance/test_tool_name_drift.py -q
# 87 passed, 1 skipped

uv run --extra dev pytest tests/test_run_agent.py -q
# 4 passed

Closes INT-530

[linear]

INT-530