sql: accept WITH (WAIT ...) on ALTER CLUSTER unconditionally

[aljoscha]

Part 4 of the design stack. Stacked on #38103.

Why

Graceful cluster reconfiguration has lived behind
enable_zero_downtime_cluster_reconfiguration, default off, so the planner
rejected any WITH (WAIT ...) clause unless a deployment turned the flag on.

That flag is now the only thing standing between an operator and the synchronous
cut-over (#38103), which is the escape hatch for reshaping a cluster when the
cluster controller itself is the problem. On any deployment running the compiled
default there would be no way to reshape a cluster at all in that situation. A
break-glass path behind a default-off flag is not a break-glass path.

What lands

• The feature_flags! entry and the require_feature_flag call in
  plan_alter_cluster. Everything else about the flag is macro-derived, so
  there is no second registry to update.
• Every test that used the surface enabled the flag itself, so those
  ALTER SYSTEM SET/RESET statements go (testdrive, sqllogictest, pg-cdc,
  test/cluster/mzcompose.py, cloudtest, parallel-workload, and the
  graceful-reconfiguration platform check).
• get_minimal_system_parameters pins the flag on below v26.38 instead. This is
  load-bearing, not tidiness: AlterClusterGracefulReconfiguration issues
  WITH (WAIT UNTIL READY ...) in its first manipulate phase, which under
  UpgradeEntireMz runs against the last released binary, and that binary still
  enforces the gate.
• The launchdarkly-flag-consistency allowlist entry moves from
  INTENTIONAL_LD_OVERRIDES to KNOWN_STALE_LD_FLAGS. The flag is a real LD
  flag, so once the last release carrying it ages out the consistency check
  would report it stale and fail. The flag should be archived in
  LaunchDarkly, after which this entry can go too.

The two rejections that share the removed code path are untouched: a WAIT
without a replica-shape change (WAIT can only be used together with a SIZE, AVAILABILITY ZONES, or INTROSPECTION change) and a WAIT on an unmanaged
cluster (ALTER... WITH not supported for unmanaged clusters). No test asserted
the gate's own error text, so nothing there changes.

enable_cluster_schedule_refresh is deliberately left alone at its current
default, even though it is the near-identical sibling gate a dozen lines below.

Reviewer attention, please

Two judgment calls worth a second opinion:

1. The private-preview badges in doc/user/content/sql/alter-cluster.md are
   removed. Nothing gates the surface after this, so describing it as private
   preview is no longer accurate. If the preview status is a product decision
   that has not been made yet, revert just that file and the rest still stands.
2. No release-note entry. doc/user/content/releases/_index.md is organized
   by released version with release dates, and v26.38 has no section yet, so
   there is nowhere to put one without inventing a date. Whoever writes the
   v26.38 notes should announce that WITH (WAIT ...) is generally available.