Auto-generate contact_ids when not explicitly set

[h-mayorquin]

Closes #423. A Probe now always carries contact_ids: when the user does not pass them into set_contacts, they are auto-generated as ["0", "1", ..., str(n - 1)], matching the zero-indexed convention and giving every contact a stable, slicing- and aggregation-invariant handle without requiring every reader or caller to generate ids themselves. This removes a class of friction where downstream code (serialization round-trips, ProbeGroup aggregation in PR #420, ChannelsAggregationRecording in SpikeInterface #4465) had to special-case None at the consumer side. set_contact_ids also accepts the legacy all-empty-strings form (produced by older serialized probes via to_numpy) and regenerates in the same way, so old data still loads cleanly. Because contact_ids is now an internal invariant after set_contacts, I deleted the unreachable None guard in the BIDS export path in io.py (which also referenced the now-irrelevant auto_generate_contact_ids workflow) and the None branch in Probe.to_numpy. I kept the change narrow: Probe.__init__ still leaves _contact_ids = None so an unpopulated probe is still recognizable, and I did not touch the manufacturer-label split (python-neo's contact_name vs contact_id); that can come later if we decide we want to preserve manufacturer labels separately from the always-set identifier.