Dashboard inference job submission

[alicup29]

Summary

Adds inference (track) job submission to the dashboard, letting users run sleap-nn inference on remote workers directly from the web UI.

Motivation

Previously the dashboard only supported training job submission. Users who wanted to run inference on trained models had to use the SLEAP GUI, sleap-app, or the CLI. This PR enables the full inference workflow from the dashboard: select models, select data, submit, and monitor — matching the training job experience.

Key Changes

Dashboard — Job type selector

• Training/Inference toggle in Step 1 of the job submission wizard
• Step indicator adapts: 3 steps for training (worker → config → files), 2 for inference (worker → files)
• Info box explains the inference flow when selected

Dashboard — Inference file selection

• Combined view for selecting model checkpoint directories and .slp data file
• Reuses the existing SSE relay file browser with inference-specific routing
• Model paths shown as cards with remove buttons, "+ Add Another Model" browse button
• "Select Folder" / "Select File" button appears on click (no double-click required)
• Submit enabled only when both models and data file are selected

Dashboard — Inference progress

• Status view shows "Running inference..." with streaming worker logs
• Hides training-specific UI (epoch counter, metrics, Stop Early button)
• Cancel button works via same ZMQ mechanism as training
• Logs saved to activeJobs for replay in Job Summary

Worker — Inference output forwarding

• Track jobs merge stderr into stdout (stderr=STDOUT) so rich/tqdm output is captured
• RelayChannel catch-all forwards unrecognized text lines and CR:: progress updates as running status messages
• All sleap-nn inference output (timing, predictions, output path) reaches the dashboard

Files changed

File	Change
dashboard/app.js	Job type toggle, inference file selection, branched submitJob(), log persistence
dashboard/index.html	Job type toggle HTML, inference file selection view, file browser container
dashboard/styles.css	Job type toggle, section labels, model cards, browse buttons, data input
sleap_rtc/worker/job_executor.py	stderr=STDOUT for track jobs, skip stderr task when merged
sleap_rtc/worker/mesh_coordinator.py	Catch-all relay for text lines and CR:: progress updates

Test plan

• Training/Inference toggle appears in Step 1
• Inference skips config upload, shows 2-step indicator
• Browse worker filesystem for model directories (Select Folder button)
• Browse worker filesystem for .slp data file (Select File button)
• Submit sends TrackJobSpec with data_path and model_paths
• Worker runs sleap-nn track and exits successfully
• Inference logs stream to dashboard worker log panel
• Job Summary shows inference logs after completion
• Cancel works during inference
• Training flow unchanged when Training toggle is selected

🤖 Generated with Claude Code