Make Trace/Stats FDSN Source Identifier–native

⚠️⚠️⚠️⚠️ this PR is heavily AI-generated, the idea is a proof of concept built on the discussion in #3559 in order to nourish our discussions & ideas ⚠️⚠️⚠️⚠️⚠️⚠️

Closes part of #3559.

What this does

Makes the FDSN Source Identifier the internal channel identity of a Trace,
while keeping the SEED channel (and NET.STA.LOC.CHA) fully working. A trace
now carries network, station, location, band, source and subsource
as its stored codes; channel is a derived view over them.

New API

• stats.band, stats.source, stats.subsource — read/write on every trace.
• stats.sid and Trace.sid — the full FDSN:… identifier (read/write).
• Stream.select(band=…, source=…, subsource=…, sid=…) — new selectors (wildcards supported).
• Stats.from_sid() / Stats.to_sid().
• Inventory.get_response() / get_channel_metadata() (and the Network equivalents) now accept a full Source Identifier as well as a SEED id.

Backward compatibility

• stats.channel, stats.component, Trace.id, Stream.select(channel=…) behave exactly as before on all existing data.
• channel is still present as a normal Stats key, so dict(stats), str.format(**stats), equality against a plain dict, and pickling are unchanged.
• Old pickles are read correctly (the stored channel is decomposed on load).
• Non-SEED and "extended" channels (e.g. long codes, multi-character or empty band/source/subsource) now round-trip losslessly, including for miniSEED 3.

Decisions worth noting

• channel stays present and stored (as a derived mirror), never removed. This keeps stats.channel access working everywhere, which was the main compatibility concern raised in Stats: How to facilitate FDSN Source Identifier? #3559 — and rules out the alternative designs that either drop channel in one mode or change the type of stats.
• A single-character channel maps to the component (subsource). So channel="Z" still gives component="Z", preserving behaviour for SOH / single-letter channels.
• Trace.id stays SEED-shaped for compatibility; Trace.sid is the unambiguous identifier.
• No new validation. Code permissiveness is unchanged; nothing that was accepted before now raises.

Internal changes

• Reader/writer and processing code that pulled band/source/subsource out of channel by position (channel[0], channel[-1], channel[:-1], …) now uses the named codes, so it is correct for multi-character codes. Migrated: io/sh, io/gcf, io/rg16, io/xseed, signal/polarization, and stream rotation/merging.

Tests

• New test module for the identifier codes, round-tripping, selection, pickling, and rotation of multi-character channels.
• Full test suite passes (only the pre-existing network-dependent doctests, which download from examples.obspy.org, fail — unrelated to this change).

Not included / follow-up

• Event metadata (WaveformStreamID in QuakeML) is not yet Source Identifier–aware. Picks and arrivals still carry only NSLC. This is independent of the waveform side and does not affect anything here; it is left for a follow-up PR.

Feedback wanted

• Agreement on the two behaviour decisions above (single-character channel → component; Trace.id stays SEED-shaped with Trace.sid added).
• Whether the follow-up for event metadata should be tracked in a new issue.

AI used?

The Claude was used heavily

PR Checklist

• Correct base branch selected? master for new features, maintenance_?.?.x for bug fixes
• Tests: Added new tests for any new features or fixed regressions
• Changelog: Added a short note in CHANGELOG.txt (only obsolete if fixing a bug introduced after the last release)
• Add the yellow ready for review label when you the PR is ready to be reviewed