Otherwise, a paragraph with multiple rows might fail to make the
scrollbar appear when displaying text.

As text is inserted, the cursor is moved to the end, and then back to
offset 0. A reordering of operations prevents a WM_NOTIFY from being
sent on cursor being moved to the end, and another by being moved
back to the beginning. If the cursor was not at offset 0, then
exactly one WM_NOTIFY must be sent, for the movement from previous
position to the beginning.  With tests to prove this change and the
previous one (modify flag should be off on WM_SETTEXT-caused
WM_NOTIFY).

Otherwise the app-defined WM_NOTIFY callback might see the modify flag
set, even though WM_SETTEXT is supposed to clear it.

EM_SETCHARFORMAT with wParam==0 sets default char format and does NOT set modify step flag.
EM_SETCHARFORMAT with wParam==SCF_SELECTION only sets modify step flag when selection is nonempty.
EM_GETMODIFY returns -1, not 1, when modify flag is set.
Tests for the above behavior.

richedit: Only notify selection change when selection has actually changed from previous notification.

Otherwise, redundant and early notifications are sent to apps that do not expect them. Fixes crash #1 with Perfect! TextEdit.

richedit: Implement rule that in 1.0 emulation, a single CR that terminates the text is NOT a line break.

Fix riched32 test that now succeed.

Document remaining uses of bEmulateVersion10 and other checks for CRLF in editor.c.
Make RTF reader emit a \r or a \r\n according to emulation, not a \n, which breaks streaming tests.
Remove todo_wine from a bunch of riched32 tests that now succeed.

richedit: Remove indication for bCRLF, now that ME_GetTextW() knows how to honor CR and LF counters.

EM_GETTEXTRANGE and EM_GETSELTEXT are affected by this, so include tests to ensure no behavior was broken.

ME_GetTextLengthEx should ignore GTL_USECRLF in 1.0 emulation mode.

Add fixup to ME_FindItemAtOffset(), fixes crash by null-pointer access.
Add tests for EM_LINELENGTH.

Add richedit 1.0 tests for EM_GETLINE.

Text streamout now honors CR and LF counters.
Tests to pin down required EM_STREAMOUT behavior.

As pointed out by Dmitry Timoshkov.