Wire up transmission segmentation in the analysis pipeline

[MuddyWinds]

Transmission segmentation helpers were added in commit 305fb18 ("feat(ingestion): add transmission segmentation helpers (not yet wired)") but are not connected to the pipeline yet. As a result, a single audio chunk that contains multiple back-to-back transmissions can be analyzed as one unit — which is a known cause of multiple aircraft being grouped into a single observation card.

Goal

Wire the existing segmentation helpers into the ingestion/analysis flow so a chunk is split into discrete transmissions before batching, and each transmission is analyzed independently.

Suggested approach

• Locate the segmentation helpers from 305fb18 and review their interface.
• Apply segmentation after STT, before the transcript enters transcript_queue (or before batch assembly in the batcher).
• Ensure each resulting segment carries its own timestamp/confidence so downstream cards and ADS-B snapshots stay aligned.
• Add a feature flag / config toggle if behavior change is risky.

Where to look

• The segmentation helpers introduced in 305fb18
• backend/core/batcher.py and the ingestion path that feeds transcript_queue

Acceptance

• A chunk containing 2+ transmissions yields 2+ separate observations rather than one merged card.
• A unit test feeding a multi-transmission transcript asserts the expected number of segments.

This maps directly to the "occasionally groups multiple aircraft into one card" limitation in the README.