see https://github.com/adsr/mle/issues/77#issuecomment-1835293952

make seems to run fine in parallel now by the way, but i'm not
removing the `NOTPARALLEL` targets just out of paranoia that it will
break a build somewhere.

when paging a full screen or greater, zero the viewport.
when paging less than a full screen, rectify the viewport.

doubt autotools is generating much beyond the generic forms of
pcre2.h and config.h on most modern systems. compiling it ourselves
is neater and requires less build deps.

`dh_auto_clean` gets confused by `make -C pcre2 clean` failing (via
output, not exit code) when you run `make clean` without submodules
checked out.

previously the condition was vwidth gt bview width. this broke in the
very specific scenario when a line was exactly as wide as the bview
with soft-wrap disabled. if you invoked `cmd_move_eol` on this line,
we would detect that the cursor is off screen and skip
`tb_set_cursor`, however viewport_x also would not apply, so the
cursor would be invisible. making viewport_x apply in the gte case
fixes this.

previously if a line's char_count equaled the buffer width exactly, and the
cursor was moved to the end of this line, we would mistakenly return an
out-of-bounds error, and not change the cursor position.

i haven't come up with a good way to handle make dependencies with
the pcre2 autotools dep. for now disable parallelism because -j >=2
breaks the build.