[cstolting-collab]

Calling .apparent() on a star with missing coordinates currently sends a NaN date to jplephem and raises a misleading EphemerisRangeError. In a catalog array, this also prevents valid stars from returning positions. This change keeps the unknown positions as NaN while valid rows continue normally.

In _compute_deflector_position(), use the observation time when the clipped closest-approach offset is NaN. The target coordinates remain untouched and propagate NaN through the deflection and aberration calculations. Finite offsets follow the same calculation as before, and real out-of-range deflector dates still raise.

The first Hipparcos example now explicitly drops rows without RA/Dec. Star.from_dataframe() continues to preserve rows and order: filtering inside that constructor would otherwise leave output coordinates misaligned with the caller's original dataframe (including the guide's magnitude-based plotting example). The guide explains how to keep unknown rows.

Validation on Linux, Python 3.12.14, NumPy 2.3.5, pandas 2.2.3, jplephem 2.24:

• Four new regression tests fail on unchanged upstream and pass with this change. They cover scalar NaN fields, scalar and array times, geocentric and surface observers, mixed/all-NaN/empty arrays, and the bundled Hipparcos excerpt. A fifth test confirms real deflector-range errors still raise.
• Full Hipparcos catalog: all 118,218 rows complete; the 263 missing-position stars remain NaN. All 117,955 valid XYZ positions are byte-identical to filtered, unmodified upstream output for both observers. Each of the 263 incomplete stars was also checked individually for both observers, with warnings and floating-point errors raised.
• test-code.sh: upstream has 1 failure out of 584 tests; this branch has the same failure out of 589 tests (test_building_time_from_list_of_utc_datetimes, an unrelated integer/float casting error). Pyflakes passes.
• The Stars page's 11 Sphinx doctests pass with the bundled data (Sphinx 9.1.1, isolated page build).

Catalog source: https://vizier.cfa.harvard.edu/ftp/cats/I/239/hip_main.dat

SHA-256: 58ceabb104d647160d9437ce6e513a02a036bb4ad9f8879a5a22fd52943616e0

Addresses the NaN-star part of #1128 and follows Brandon's preference in #1128 (comment). This builds on the diagnosis by ajorennie and karimosama7. The separate rise/set issue remains covered by #1145.

Reproduction scripts, catalog evidence, and the separately scoped almanac follow-up are available in the published review package. The CML review harness is optional and adds no dependency to Skyfield.

Prepared and tested with Codex assistance.