Convert timetables, like tables

[Thenewmanator15]

A MATLAB timetable is read as a MatlabOpaque: its parts (variable names, data, row times, units) are all decoded, but never assembled. This converts it the way table already is — to MatlabTable by default, or to the table type (DataFrame, ...) — with the row times as the first column, named after the row dimension (Time unless renamed).

vars = matread("timetable.mat"; table = DataFrame)
vars["tt_datetime"]
# 3×3 DataFrame
#  Row │ Time                     Current   Channel
#      │ DateTime                 Float64   String

Row times, in the forms MATLAB saves them:

• one per row, as datetime or duration — already converted to DateTime / Millisecond by the time the timetable is;
• regular timetables (SampleRate or TimeStep), which MATLAB stores as a start time plus a rate or step — the row times are generated;
• a step in calendar units (calmonths, ...) has no fixed rate: the timetable is left as the MatlabOpaque, with a warning.

table = Nothing keeps the raw object, as for tables.

Tests, and a request. test/timetable_gen.m writes test/v7/timetable.mat and test/v7.3/timetable.mat with one variable per case — datetime and duration row times (sub-millisecond ones included), sample rate, time step from a datetime start, an empty timetable, a two-column variable, a renamed row dimension, and calendar months. I can't contribute the .mat files themselves, so they are not in this PR: could someone with MATLAB run the script in test/ and commit its two files? Until then the new testset skips (@test_skip), so CI stays green. Run locally against the files the script writes (MATLAB R2025a), all 48 new assertions pass and the full suite passes (629) on Julia 1.13; read.jl also passes on 1.9. Without the files: 581 pass, 2 skipped.

I also checked it on a real acquisition file (v7.3, a user-defined class holding four timetables, one of them empty): every value matches two other readers of the same file, and the row times agree to within DateTime's millisecond resolution.

A note on ms_to_datetime, which asks whether the imaginary part of a stored datetime is the sub-millisecond part: it is the low half of a double-double, below the resolution of the real part (on that file, |imag| ≤ half an ulp of the real part in every row), so real + imag in Float64 equals real and dropping it loses nothing at DateTime resolution. The sub-millisecond detail is the real part's own fraction, which DateTime rounds away.

Independent of #251 (no overlapping lines); docs: a timetable row in docs/src/types.md and the matopen / matread docstrings.

🤖 Generated with Claude Code