[Sad Catapella N5][Data integrity] Header-only schema-v4 CSV is accepted as an empty experiment #48

Baseline: f784c95. Confirmed API inconsistency, no package tests run.

Code: Writer refuses to publish an empty experiment. Reader validates a v4 header, then returns [] when no data rows follow. It checks a completely empty file, but not a valid header-only file. Tests specifically reject a header-only v3 file without covering v4.

Impact: A truncated or externally produced v4 file with zero planes is treated as successfully parsed; downstream failure appears far from the corrupted file. No claim that the writer itself produces this state under normal atomic publication.

Fix: Reject a header-only valid v4 file with CsvFormatError naming the missing XYC plane rows and recommending regeneration. Regression: valid-v4-header-only, completely empty file, single valid row and legacy-version header-only, preserving migration messages. No Numba benefit.