Add warning if an empty line occurs and `max_rows` is used

[seberg]

This wraning may just be around for a few releases, but to inform users that whatever they had may have gotten broken... Hopefully, almost nobody will see the warning anyway.

[seberg]

OK, I added a super-verbose warning right now (which can probably be improved though). It is unclear whether we may want to remove this eventually, etc.

One test is definitely missing: We need to a test that uses an actual file (as this hits the chunked reader path), it is not impossible there is a bug there right now.

The current warning is:

In [3]: np.loadtxt(["asdf\n", "1,2\n", "\n", "3,4\n"], dtype=int, skiprows=1, delimiter=",", max_rows=2)
<ipython-input-3-b783576405dc>:1: UserWarning: Input line 3 contained no data and will not be counted towards `max_rows=2`.  This differs from the behaviour in NumPy <=1.22 which counted lines rather than rows.  If desired, the previous behaviour can be achieved by using `itertools.islice`.
Please see the 1.23 release notes for an example on how to do this.  If you wish to ignore this warning, use `warnings.filterwarnings`.  This warning is expected to be removed in the future and is given only once per `loadtxt` call.
  np.loadtxt(["asdf\n", "1,2\n", "\n", "3,4\n"], dtype=int, skiprows=1, delimiter=",", max_rows=2)

[seberg]

Also forgot to fix the release notes right now.