This solves the problem of rows with too big minimal height causing
layout failure (the table just goes out of page, does not flow to
next page).

It does so with three steps:
1. Currently, a row with a minimum height that flows to next page
repeats whole min height on that (and possibly following) pages.
If this behaviour continued, then that would cause layout loop:
the row min height would be too high for the page, so it would
keep flowing to next pages (actually just go beyond the botom
because of layout failure). To mitigate this, the patch changes
the behaviour to measure total height of all frames of the row:
the function lcl_calcHeightOfRowBeforeThisFrame calculates the
total height of previous row frames for the same row, then in
places where we need to get min height, this value is subtracted
from row min total height. On following pages the min height of
frames will get less each time.

2. When the row is split, the possibility to split depends on if
the minimum height of the row fits into the vertical space left.
The minimum height is found as maxinum of two values: minimal
contents of the row (e.g., height of first line of text, or an
object, or lower table cell), and the minimum height setting.
As the minimum height setting should not prevent the cell to
flow, (it only should ensure that total height is no less), we
should not consider the setting when the split is performed
(we should be able to keep on first page as little as required).
To allow this, a new bool member introduced in SwRowFrame:
m_bIsInSplit, with its setter and getter. When it is true,
the routines that calculate min height ignore the min height
setting. It is set in lcl_RecalcSplitLine around lcl_RecalcRow
and SwRowFrame::Calc that decide if it's possible to keep part of
row's content on first page, and update table's height to fit
the rest of space.

3. It turns out, that if contents of the splitted cell has less
height than the min height setting, then following happens.
In SwTabFrame::Split, all rows below splitted one are moved to
follow flow table; then table frame's Shrink method used to shrink
the freed height. At this moment, still, the height of splitted
row is too high, and total height of table is too high. Then,
lcl_RecalcSplitLine is called, where row is first shrunk, and then
lcl_RecalcRow and SwRowFrame::Calc try to move contents and resize
the splitted row. They get the minimum height of content (we already
don't consider the min height setting), but then "last row will fill
the space in its upper" (see SwRowFrame::Format). Row returns its
previous size, table does not resize, it doesn't fit to page, and
split fails.
To try to fix that, I call SwTabFrame::Shrink in lcl_RecalcSplitLine
after lcl_ShrinkCellsAndAllContent before lcl_RecalcRow.

Unit test included.

Change-Id: I8422e43cff7d6b5955541ed3fe930779429ed55b
Reviewed-on: https://gerrit.libreoffice.org/31774Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Miklos Vajna <vmiklos@collabora.co.uk>