fix: Handle multi-output fan-out in _build_output_mapping_from_edges (langflow-stepflow)

[coderabbitai]

Summary

In src/langflow-stepflow/src/langflow_stepflow/translation/translator.py, the _build_output_mapping_from_edges method uses a first-edge-wins strategy that collapses all outgoing edges from a single source node into a single source_id -> output_name entry. This is lossy when a node fans out to multiple targets using different outputs — at least one downstream path will receive the wrong output.

Steps to reproduce / Follow-up work

1. Identify (or author) a Langflow flow where a single component exposes multiple outputs and connects each to a different downstream node (e.g., a component with both a message and a data output, each wired to a separate target).
2. Verify that the current first-edge-wins mapping causes the wrong output to be selected for at least one consumer.
3. Change the mapping to be edge-scoped (e.g., keyed by (source_id, target_id) or by edge id) so each edge retains its selected output name.
4. Update NodeProcessor (and any other callers of _build_output_mapping_from_edges) to look up the output using the edge-scoped key.
5. Add a regression test using the flow from step 1.

Context

Flagged during review of PR #12015. The first-edge-wins approach works for all existing tested flows, but is architecturally unsound for general multi-output fan-out scenarios.

Backlinks: PR #12015, comment #12015 (comment)

Requested by @bjchambers.