[github-actions]

Merging this PR tags v0.13.0 and publishes it.

Bump: minor, from the commit titles since the last release.
To ship more first, keep merging; this PR updates itself.
Polish the notes by editing CHANGELOG.md here, just before merging:
the next merge to main rebuilds this branch and drops edits made here.

Release notes

Added

• tx deploy checks the deployed model after it lands: it re-reads the target and reports every
  object the server holds in an error, incomplete, or unprocessed state, in the text output and
  as validation in the JSON result. Objects in an error state (a measure whose DAX does not
  resolve, a broken RLS filter) fail the deploy with exit code 1 and
  TOMIX_DEPLOY_VALIDATION_FAILED, even though the model is deployed; incomplete objects warn
  (TOMIX_DEPLOY_VALIDATION_WARNINGS); unprocessed ones point at tx refresh. A check that
  cannot run warns with TOMIX_DEPLOY_VALIDATION_UNAVAILABLE. --ci vsts|github now prints
  annotations for post-deploy errors and warnings.
• tx deploy --rollback-on-error checks the deployed model before the deploy is kept. On an
  existing model the deploy runs inside a server transaction and rolls back when an object is in
  an error state, leaving the target's definition and data unchanged. A new model is deleted
  again. A rolled-back deploy fails with TOMIX_DEPLOY_ROLLED_BACK and "status": "rolled-back".
  See ADR 0005.
• tx format style options (feat(format): shared format options (--semicolons, skip-space-after-function) + output goldens #229): --separators us|eu shows DAX with ; list and ,
  decimal separators (display only; models always store the US form, and an explicit
  --separators eu with --save, --save-to, --stage or --revert fails with the new
  TOMIX_FORMAT_SEPARATORS_DISPLAY_ONLY), --function-spacing space|none drops the space
  before a DAX call's parenthesis, and --short undoes --long. Each has a config key
  (format.separators, format.functionSpacing, format.lines) that the flag overrides.
  Golden files pin the printed output for each style.
• tx bpa run --vpax <file.vpax> reads VertiPaq statistics from a file exported with
  tx vertipaq --export, so statistics rules run on a model file, a read-only connection, or in
  CI without annotating the model. The statistics apply in memory only. A file that matches no
  table fails the run (TOMIX_BPA_VPAX_MISMATCH); tables missing from the model warn
  (TOMIX_BPA_VPAX_PARTIAL) (feat(bpa): --vpax input and default-off size-stats rules #234).
• New default-off vertipaq ruleset preset with every bundled rule that reads statistics,
  including two new rules: LARGE_SHARE_OF_MODEL_SIZE and
  UNUSED_HIDDEN_HIGH_CARDINALITY_COLUMNS. Rules can read model-level annotations through
  Model.GetAnnotation(...) (feat(bpa): --vpax input and default-off size-stats rules #234).

Changed

• FIX_REFERENTIAL_INTEGRITY_VIOLATIONS, AVOID_BI-DIRECTIONAL_RELATIONSHIPS_AGAINST_HIGH-CARDINALITY_COLUMNS
  and LARGE_TABLES_SHOULD_BE_PARTITIONED move to the vertipaq preset (category VertiPaq),
  out of standard and full. bpa run without --ruleset and the deploy BPA gate add the
  vertipaq preset when the model has statistics (Vertipaq_* annotations or --vpax), so an
  unannotated model no longer lists them as "Not checked". A named --ruleset is used as given:
  --ruleset standard now skips them (feat(bpa): --vpax input and default-off size-stats rules #234).

Fixed

• tx test --tolerance (and tx: tolerance=) now applies to measures whose data type is Variant.
  Their results read back as object columns, which compared exactly, so floating-point noise
  failed the test whatever the tolerance. Cells compare within the tolerance when both sides are
  numbers; text, dates and booleans still compare exactly. --ignore-row-order sorts these
  columns by value after the dimension columns, in a fixed BLANK, number, text order.