Summary of important user-visible changes for datatypes 1.5.0:
--------------------------------------------------------------

 This release adds the `timeseries` and `tscollection` classes, with
 `timeseries2timetable` to carry them onto a timetable, and aligns the
 conversion and joining of table variables with MATLAB.  It breaks code
 that relied on the old conversions; see below.

 ** New classes: `timeseries`, samples on a time vector with units, a start
    date, quality codes and events; `tscollection`, several series sharing
    one time vector; and their metadata classes `tsdata.timemetadata`,
    `tsdata.datametadata`, `tsdata.qualmetadata`, `tsdata.interpolation` and
    `tsdata.event`.

 ** `timeseries` has every method of MATLAB's but `plot`: selecting and
    editing samples, dates, the six event queries, `resample` and
    `synchronize`, nine statistics with quality, missing-data and time
    weighting, the arithmetic operators sample by sample, `detrend`,
    `filter` and `idealfilter`.

 ** `tscollection` has every method of MATLAB's: adding, removing and
    renaming members, editing samples, dates, `resample`, and concatenation
    in time and across members.

 ** `timeseries2timetable` converts series, or the members of a collection,
    into a timetable, with their units, continuity and events.

 ** New functions `istimeseries`, `istscollection` and `strings`.
    `table2array`, `table2cell` and `table2struct` take a timetable.

 ** Table display shows integers exactly, both parts of complex numbers and
    the row times of a nested timetable.  `standardizeMissing` takes
    `duration` and `datetime` indicators.  Unknown Name-Value options are
    reported in one wording across the package.

 ** The Time Zone Database is updated to release 2026e: Manitoba keeps
    UTC-5 all year from November 2026.

 ** 33 bug fixes, the largest groups in set operations and joins on rows
    with missing values, nested table variables, and table display.

 ** Deliberate deviations from MATLAB are documented in the docstrings of
    the functions they affect.  The ones most likely to be met: `setabstime`
    keeps each sample with the date it is given, where MATLAB sorts the
    dates alone; time weighting gives true time-weighted medians, spreads
    and extremes, where MATLAB's are right for `mean` and `sum` only;
    `timeseries2timetable` keeps the events MATLAB drops; `tscollection`
    member names are unique ignoring case and read exactly.


Backwards incompatible improvements:

 ** `table2cell` and `table2struct` keep each variable's type, where
    `categorical`, `datetime`, `duration`, `calendarDuration` and `string`
    variables gave text.

 ** `table2array`, `{}` indexing and `.Variables` join the variables as
    MATLAB does, and return a cell array where they cannot form one array,
    where they raised.

 ** `{}` assignment copies a one-row value down the rows only when it
    creates a single new variable, as MATLAB does.

 ** Concatenating a `categorical` array takes only text, missing values and
    empty arrays beside it; `[c; 1]` added a category `'1'`.

 ** `sortrows`, `topkrows` and `issortedrows` refuse a `calendarDuration`
    sort key, as MATLAB does.

 ** `ismissing`, `rmmissing` and `anymissing` never count a nested table
    variable as missing, as MATLAB does.

 ** `length` on a table raises an error pointing to `height`, `width` and
    `size`, as on a timetable.
