fix: omit transport from dataplane config

[lucarlig]

Summary

• keep the dataplane publisher limited to STREAMABLEHTTP gateways
• omit the obsolete transport property from published backend configuration
• cover both SSE and STDIO with the existing non-Streamable gateway filter test

Why

The Rust dataplane now has a single unconditional upstream transport and no
transport selector. The control-plane publisher should match that contract
instead of sending a field the dataplane no longer models.

This is the control-plane companion to
contextforge-org/contextforge-data-plane#79.

Impact

SSE and STDIO remain available on control-plane-owned routes. This change only
prevents those gateway types from entering dataplane configuration and removes
transport selection from the published backend payload.

Validation

• dataplane publisher unit tests: 25 passed
• Ruff: passed
• Bandit: passed
• pylint on the publisher: 10.00/10
• repository pre-commit hooks: passed

[cafalchio]

I think it should filter to send the compatible gateways, otherwise we will need the transport to filter it out when calling a tool in dataplane.

[Lang-Akshay]

LGTM

[msureshkumar88]

Reviewed the change against the companion Rust PR (contextforge-data-plane#79) and the existing filter logic — this is a clean, minimal fix.

What it does: drops the now-unused transport field from the published BackendConfig/dataplane payload, matching the Rust side's move to a single unconditional upstream transport. The STREAMABLEHTTP-only gateway filter itself is unchanged (pre-existing logic, confirmed via blame) — only the comment was reworded and the field removed from the outbound payload.

Verified:

• No other in-repo consumer references the removed payload key.
• DbGateway.transport is still read internally (needed for the filter) and just no longer serialized — not dead code.
• Test changes are in-scope: dict-equality assertion now guards against the field silently reappearing, and the non-streamable-gateway test is parametrized to cover both SSE and STDIO.
• Ran the full test_dataplane_publisher.py suite in an isolated worktree: 25/25 passing.
• No UI, alembic/schema, or docs changes needed — none touched, correctly.

One non-blocking note: since this changes a cross-repo wire contract with no schema version field, an environment that already has dataplane_publisher=true enabled would break if this lands before the companion data-plane#79 deploys. Worth a one-line rollout note (deploy dataplane side first), but given the flag defaults to false and the module is explicitly marked experimental, this isn't a blocker.

Approving.