Make DOM data attribute loading fully lazy when counters are available

* If the input pagebundle provides counters, there is no reason to
  eagerly walk the DOM to initialize annotation and about id counters.
  Effectively, prepareAndLoadDoc skips visitAndLoadDataAttributes
  altogether.

* In practice, for OutputTransformPipeline passes in core, if the passes
  never need to inspect a node's data-mw or data-parsoid attributes,
  the node's data is left behind in the input pagebundle without
  being loaded at all.

  However, this now requires that storeRichAttributes properly transfers
  the node attributes either to the output pagebundle or to inline
  attributes (depending on the output DOM format) from the input
  pagebundle or inline attributes.

  In practice, for input and output formats relying on inline attributes,
  this becomes a noop. And, for input and output formats that
  use pagebundles, this requires a simple copy from the input pagebundle
  to the output pagebundle which is much cheaper than the full
  load-parse-serialize-store operation.

  But for PB -> inline or inline -> PB transfers, the benefits from
  lazy loading are lesser but still present.

* Effectively, this patch eliminates the "load" part of the
  prepareAndLoadDoc method.

  This patch removes all unnecessary calls to visitAndLoadDataAttributes
  (and adjusted couple test files to account for the lazy loadng)

  At this time, there is one remaining call to this method in ContentUtils
  which we should revisit and see if it can be handled differently.

* Updated existing lazy-loading unit tests to deal with additional
  laziness from this patch.

* Added an additional test to test precedence of inline attributes
  over pagebundle attributes.

* With these changes, all parser tests & unit tests pass.

Bug: T409751
Co-Authored-by: C. Scott Ananian <cananian@wikimedia.org>
Change-Id: I41e6d4e98e123f68e14ffe7198e5015b36b75a37

Revert "Fix a corner case in table attribute parsing"

This reverts commit 582d2b2.

Reason for revert: We want to tag a version of Parsoid to deploy
as a backport and this patch hasn't been rt-tested but patches on
top of it have been. So, temporarily reverting to get us a taggable
version of Parsoid for deploy.

Change-Id: I71c42a5c38e1cae2a2a310d07e81aa166df62be2

Sync parserTests with TMH and core

No actual test changes, just updating hashes prior to the REL1_46
branch point.

Change-Id: I7da91bdf076f3ac3b1e1c890fc24545d8392f29d

Sync parserTests with TMH and core

No actual test changes, just updating hashes prior to the REL1_46
branch point.

Change-Id: I7da91bdf076f3ac3b1e1c890fc24545d8392f29d

Sync parserTests with core

Some newly-added parser tests related to core's implementation of
LanguageConverter were added to the known failures list; these
should probably not have had the `wt2wt` mode enabled in the first
place.

Change-Id: Id36d3ecc2490edc3901bc64597898842a3cf9e64

Update phpunit to 10.5.63

Bug: T423763
Change-Id: If20c736156625e5d57bb034b96cbe2e2d893f582

Don't language-convert inside <style> elements

Bug: T422879
Change-Id: I67538c64168ea6447aab6f63bd836307bf948666

Don't language-convert inside <style> elements

Bug: T422879
Change-Id: I67538c64168ea6447aab6f63bd836307bf948666

Don't language-convert inside <style> elements

Bug: T422879
Change-Id: I67538c64168ea6447aab6f63bd836307bf948666

Skip performance test that crashes pcov so that coverage reports don'…

…t fail

Bug: T422865
Change-Id: If60713623348288f5cf89592390234e0a189b498