Integrate or remove the /api/split_inference endpoint

[ff225]

Priority: High

The /api/split_inference endpoint (~160 lines) is a prototype wired directly into the HTTP server and disconnected from the main pipeline:

• hardcoded skip connections in a dict ({25:[16,24], 43:[34,42], 52:[43,51]})
• hardcoded model folder names (FOMO_48_CUT, FOMO_96_CUT, …)
• fragile size parsing: device_id.split("_")[1].split("x")[0]
• does not use RequestHandler, OffloadingAlgo, Edge.run_inference, or ModelManager
• not covered by the evaluation/telemetry system, so it can diverge from the main flow

Decision

Either integrate it into the main pipeline (reusing ModelManager/RequestHandler) or remove it.

Files

• src/server/communication/http_server.py

From the Feature 042 codebase review (recommendation #4).