Count last-item space toward line width in more cases
Normally, when Pango wraps a line after a space character, the space
character itself is said to be collapsed, and is not counted towards
the width of the line. This is implemented by adding a negative width
adjustment (extra_width). This has been implemented in commit
86303ad2
"Handle break-after-space correctly", and a few follow-up fix-ups.
There is a corner case: when the space we're trying to break at is the last character in the paragraph, it should not get collapsed. This has been implemented in commit 8ba98271 "layout: Fix a cornercase of line width accounting". However, due to parts of the logic being duplicated in different places, that patch forgot to update some of those places. Fix that by not collapsing the space width in the other parts of the logic too.
In practice, this manifested as a Pango layout reporting to be wider
than its explicit width limit (layout->width), because the final space
was counted as collapsed (negative extra_width) when deciding whether to
break the line, and as uncollapsed when measuring the resulting line
width afterwards. That was in turn causing issues in GTK, where a
GtkLabel widget would get confused because of this fact, and report an
incorrect minimum width, and that, in turn, would be breaking
assumptions in the layout machinery several layers up in the widget
tree.
Partially un-breaks Tuba with gtk!6569 applied. Test (adapted from https://lea.pet/notes/9lc92xktj5dq3w02) included.