adapter: don't leak arrangement compression into the realized config

[aljoscha]

Stacked on #38104.

Arrangement compression is a replica config shape dimension: it is part of ReconfigurationTarget, the reconcile kernel distinguishes replicas by it, and the controller transitions it through the hydrate-overlap like SIZE. Several hand-maintained dimension lists forgot it.

The reshape leak. reshape_alter_cluster_managed writes the reconfiguration record while resetting the config shape carried by new_config back to the realized values, deferring the transition to the record's cut-over. The reset enumerated size, replication factor, availability zones, and logging, but missed arrangement compression. A compression-carrying ALTER thus flipped the realized config immediately: the baseline strategy then desired the new shape and bounced the baseline replicas mid-reconfiguration, defeating the graceful overlap, and the cut-over later wrote the same value again. The reset now goes through realized_reconfiguration_target / apply_reconfiguration_target instead of hand-listing dimensions, so a target dimension can no longer be missed at this site.

The planner gate. The WAIT gate rejected a compression-only ALTER as having nothing to wait on, even though such an ALTER transitions through a hydrate-overlap like any other shape change. It is now allowed (and the error message names the option).

The observability surface. mz_cluster_reconfigurations.changes diffed only the other four dimensions, so a compression-only record reported an empty diff while in progress, and the SHOW CLUSTERS activity summary showed such a cluster as steady mid-reconfiguration. Both now report the dimension.

Tests. The compression ALTERs in managed_cluster.slt relied on the leak: their SHOW CREATE readback only saw the new value because the reshape wrote it early. They now request the synchronous cut-over (WAIT FOR '0s'), which advances the realized config in the ALTER's own transaction. ALTER ... RESET takes no WITH clause, so the RESET spelling is exercised for acceptance without a readback. cluster-controller.td gains regression coverage: the cc_preserve section asserts that a folding compression ALTER lands in the record's target and changes while the realized config stays untouched and only the overlap replica is bounced, and a new section runs a compression-only ALTER through the record path end to end.

Known gap, left for a follow-up: the audit event's reconfiguration details intentionally omit compression, and a shape-dimension RESET cannot carry a WAIT clause (the parser only accepts WITH on SET).

Motivation

Fixes a bug in a feature that has yet to be released.

Checklist

• This PR has adequate test coverage / QA involvement has been duly considered. (trigger-ci for additional test/nightly runs)
• This PR has an associated up-to-date design doc, is a design doc (template), or is sufficiently small to not require a design.
• If this PR evolves an existing $T ⇔ Proto$T mapping (possibly in a backwards-incompatible way), then it is tagged with a T-proto label.
• If this PR will require changes to cloud orchestration or tests, there is a companion cloud PR to account for those changes that is tagged with the release-blocker label (example).
• If this PR includes major user-facing behavior changes, I have pinged the relevant PM to schedule a changelog post.