Fix AutoPlatepar CALSTARS requirement and misleading success message

[Cybis320]

• Auto-generate CALSTARS file when missing (matching SkyFit2 behavior)
• Return None instead of unvalidated intermediate platepar when star matching fails, preventing misleading "SUCCESS" after "ERROR"

[Copilot]

Pull request overview

This PR updates AutoPlatepar.autoFitPlatepar() to better match SkyFit2 behavior by auto-generating a CALSTARS file when one is missing, and by returning None (instead of an intermediate/unvalidated platepar) when star matching/final fit fails to avoid misleading “SUCCESS” outcomes.

Changes:

• Auto-generate CALSTARS via extractStarsAndSave when no CALSTARS file exists in the directory.
• Adjust verbose output to distinguish between loaded vs generated CALSTARS.
• Change several failure returns to return None for the platepar (and currently None for matched stars) rather than returning an intermediate result.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Cybis320]

I've replaced the brute-force matchStars in CyFunctions.pyx with a scipy.spatial.cKDTree implementation. The new version lives in RMS/Astrometry/MatchStars.py and is a drop-in replacement with the same signature and return format.

[dvida]

Is the new approach really faster? Building KD trees takes time and by definition querying is faster only when done multiple times. Here, the KD tree is built for each run.

[Cybis320]

Is the new approach really faster? Building KD trees takes time and by definition querying is faster only when done multiple times. Here, the KD tree is built for each run.

 ┌───────────────────────┬──────────┬──────────┬─────────────────────┐
 │       Scenario        │  Cython  │ KD-tree  │       Winner        │
 ├───────────────────────┼──────────┼──────────┼─────────────────────┤
 │ Small (20×50)         │ 0.002 ms │ 0.035 ms │ Cython 18x faster   │
 ├───────────────────────┼──────────┼──────────┼─────────────────────┤
 │ Typical (200×500)     │ 0.147 ms │ 0.120 ms │ ~tie                │
 ├───────────────────────┼──────────┼──────────┼─────────────────────┤
 │ Large (500×2000)      │ 1.54 ms  │ 0.46 ms  │ KD-tree 3.3x faster │
 ├───────────────────────┼──────────┼──────────┼─────────────────────┤
 │ Dave 360 (2000×10000) │ 28.9 ms  │ 2.9 ms   │ KD-tree 10x faster  │
 └───────────────────────┴──────────┴──────────┴─────────────────────┘

Cython wins for very small inputs (20×50), that's 20 detected stars x 50 catalog stars, but those are 0.002ms - nobody will notice, and that's not a typical scenario. At typical RMS sizes it's a wash, and it scales much better for larger inputs.

matchStars works in pixel space, so the pixel positions change every time and the KD-tree has to be rebuilt for each frame.