Avoid duplicate Iceberg asset runs after a triggerer restart

[1fanwang]

Rationale for this change

Two assets watch one Iceberg table through one trigger. Restart the triggerer for a deploy and both fire again for a commit already reported, so every DAG watching that table gets a duplicate run.

The watermark stops that, and the trigger drops it here. It also detects the case by matching an exception's text against "concrete inlets and outlets", wording from a private helper that nobody promised, so a reword turns the guard into a re-raise. #71460 makes the accessors iterable, so the trigger writes to each asset.

What changes are included in this PR?

One cursor per asset, not one or none:

stored = [a.get(WATERMARK_KEY) for a in accessors]   # oldest wins, so a partial
...                        # write repeats a snapshot rather than skipping one
for accessor in accessors:
    accessor.set(WATERMARK_KEY, head)

A state store failure now propagates, no longer read as the several-assets case. Before #71460 the accessors are neither iterable nor countable, so the trigger reads once to see if it can address them, polling unwatermarked if not.

Are these changes tested?

Live, on Postgres with a real Iceberg catalog and two assets on one trigger row.

Restart with no new commit, before and after

# append a row to sales.orders via pyiceberg, then:
$ kill $(pgrep -f "airflow triggerer") && airflow triggerer &  # no new commit
$ psql -c "SELECT e.timestamp, e.extra::json->'payload'->>'snapshot_id' snap,
  e.extra::json->'payload'->>'previous_snapshot_id' prev FROM asset_event e
  JOIN asset a ON a.id=e.asset_id WHERE a.name='orders_raw' ORDER BY 1"

--- on main ---   (orders_reporting identical)
 00:08:58 | 9203870129439515238 | 5620644195193149555  # the commit
 00:10:08 | 9203870129439515238 |                      # RESTART: again

--- with this change ---
 00:11:56 | 9203870129439515238 |     # RESTART after this: no event
 00:14:18 | 9209950183446620906 | 9203870129439515238

$ pytest providers/apache/iceberg/tests/unit/apache/iceberg/triggers/ -q
5 failed, 18 passed  # on main
23 passed            # with this change

That last prev proves the watermark survived. A stand-in for the pre-#71460 accessors covers the older path.

Are there any user-facing changes?

A table watched by several assets keeps its watermark, so restarting the triggerer no longer fires a duplicate run for a handled commit. One asset behaves as before. Stacked on #71460.

[1fanwang]

cc @vincbeck as a follow up to #71387 and #71460, thank you!