Restore full Vetstar feature set to VetStar Pi (minus multi-sector + …

[eagnespuerto]

…joint fit)

Vendors the complete backend/app package into pios/vetstar_pi so the Pi port now drives the same code paths as the cloud studio, dropping only the two features that don't belong on a Pi:

• Multi-sector representative-picker: two-pass "fetch every sector, score, re-run the winner" loop keeps too much per-sector state live to fit the 1 GB budget. The multi-sector orchestration in pipeline.py and the multi-sector PDF builder in report.py are excised.
• Joint TESS + Gaia PSPL fit: research-tier online-only workflow from Harris+2026 that only makes sense when the joint pipeline is running continuously. fit_pspl_joint and analyze_event_joint stripped from microlensing.py. The Gaia Alerts fetcher stays — the GUI's Classifier tab loads Gaia-band curves directly for the single-band fit.

Everything else in the cloud studio is now here and drives the same functions:

• MAST SPOC light-curve fetch by (TIC, sector) + sector listing
• SPOC DVT phase-fold + fitted a/R★ + impact parameter ingestion
• Adaptive event detection + gap handling
• Optional sinusoidal pre-BLS detrend (spotted rotators)
• Odd/even, secondary, centroid, shape, physics, verdict
• Constrained BLS (known-period ± 2%) with P/2 and 2P harmonics
• Gaia DR3 + SIMBAD + NASA Exoplanet Archive cross-match
• HCI habitability engine (STEHM) + subscore breakdown
• Predicted Observables (POE) + ExoFOP TOI parameter table
• TLCM transit geometry + absolute masses
• ExoMiner feature/view extraction (matplotlib, no TF)
• TESScut FFI cutouts (transit) and TESScut + Gaia overlay (microlens)
• Microlensing coverage via tess-point sector lookup
• Gaia Alerts G-band fetcher + cone search
• Stage-by-stage progress reporter wired to a Tk progress bar
• Full transit + microlensing PDF reports with DejaVu Sans

GUI (Tkinter) now exposes:

• Transit tab: Open FITS / Fetch from MAST… / List sectors…, six sensitivity knobs (threshold, min SNR, secondary σ, odd/even σ, rotation period, known period) plus high-variability detrend, four PDF-extras toggles (HCI, ExoMiner, FFI, DVT), stage/percent progress bar, embedded matplotlib canvas (LC + zoom of deepest event), verdict text pane, Save PNG / PDF / JSON. Known-object cross-match hit + HCI headline surface in the text pane.
• Microlensing tab: two sub-tabs.
  • Classifier: Open CSV/JSON + Fetch Gaia Alert…, window entry, 3-way BIC fit, embedded plot with model overlays + BIC bar chart, full param + observable pane, PNG + PDF export.
  • Coverage: Load events CSV → tess-point per row → sortable Treeview with observability flag + ecliptic latitude + bulge blind-zone flag, observability summary bar chart, Save JSON.

CLI adds sub-commands mirroring the cloud API's non-multi-sector, non-joint-fit endpoints: mast, sectors, coverage, habitability, exominer, ffi, alerts. Everything fails soft when offline; the local transit pipeline still runs.

Requirements pull astroquery (Gaia/SIMBAD/MAST/TESScut/Vizier) and tess-point (coverage). Both ship as piwheels-prebuilt ARM wheels; install stays under 5 min on a 1 GB Pi. Peak RSS with the full PDF (HCI + ExoMiner + FFI) is ~420 MB — under half the budget. The README's feature comparison table and pios/README.md are rewritten to match the new scope.