Improve A11ySnapshotTransformer to match Talkback behaviour on PW

[luizgrp]

It has a conditional to display only content description when the text semantics is present:

horologist/roboscreenshots/src/main/java/com/google/android/horologist/screenshots/a11y/A11ySnapshotTransformer.kt

Lines 129 to 133 in bccd491

	if (it.contentDescription != null) {
	drawItem("Content Description \"${it.contentDescription.joinToString(", ")}\"")
	} else if (it.text != null) {
	drawItem("Text \"${it.text.joinToString(", ")}\"")
	}

However, as per #1920 (comment), it seems that either compose or talkback are treating nodes with mergeDescendants = true as separated nodes.

It looks like the elements fetched by A11ySnapshotTransformer should be retrieved with mergeDescendants = false to match the behaviour. The side effect is that the screenshots won't "group" the a11y text boxes with the merged descendants.

[yschimke]

Partially addressed (against Gustavo's better judgement) in https://github.com/google/horologist/pull/1922/files#diff-d2295c73b71d48e87ce29bdfd15e7c7782a6f1f13e8732d9815034093ec8faa4