tm state: don't populate metadata in updateLocked

[deepthi]

Description

This is a partial revert of #8107. Calling populateMetadata in updateLocked causes publishState to fail when semi-sync is enabled.

Related Issue(s)

Fixes #8505

Checklist

• Tests were added or are not required
• Documentation was added or is not required

Deployment Notes

[deepthi]

@ajm188 it's possible that the right fix is as simple as changing these conditions:
https://github.com/vitessio/vitess/pull/8107/files#diff-8b21ba72e9fc7850bc4516d90ebf0639ac39fbef7e7b0ac534e2f470574c5a77R290-R297

	if ts.tm.MetadataManager == nil {
		return
	}

	if ts.isOpening && !*initPopulateMetadata {
		return
	}

During PRS isOpening is false and the code gets executed even when initPopulateMetadata is false. I believe the check really needs to be

	if ts.tm.MetadataManager == nil || !*initPopulateMetadata {
		return
	}

[deepthi]

@ajm188 it's possible that the right fix is as simple as changing these conditions:
https://github.com/vitessio/vitess/pull/8107/files#diff-8b21ba72e9fc7850bc4516d90ebf0639ac39fbef7e7b0ac534e2f470574c5a77R290-R297

On second thought, populateMetadata should NEVER be called in a code path that runs during a reparent.
No DB writes can happen during that process because they will hang when semi-sync is on.

[ajm188]

I am thinking we should add a flag to the section of ChangeTabletType here, and then only call populateMetadata if we're not transition from/to MASTER

[deepthi]

I'm merging this for now to fix the regression. It will need to be fixed properly later (with appropriate tests).