Default missing firstSample to 0 in SpikeGLX reader

[h-mayorquin]

Closes #1819. Some SpikeGLX .meta files do not carry firstSample, even though it is documented in every phase. The reader accesses it directly in _parse_header and crashes with a bare KeyError. This PR makes the read defensive and defaults to 0.0 with a UserWarning naming the file. Zero is the correct fallback: firstSample exists to offset t_start for files that represent a trimmed slice of a longer run, and files that never had it written are not trimmed.

I have lifted the read into a new pre-dict step, _add_first_sample, that follows the pattern of _add_segment_order: it enriches each info dict with info["first_sample"] before signals_info_dict is built, and the timing loop then reads the prepared field instead of touching the raw meta.