fix: stop ASCII subplot save from segfaulting (#2019)

[krystophny]

Closes #2019.

Problem

Saving a figure that uses a subplot grid to the ASCII backend
(savefig('plot.txt')) crashed with a segmentation fault. The same figure
saved fine to PNG/PDF.

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
#3  __fortplot_subplot_rendering_MOD_render_subplots
#4  __fortplot_figure_render_engine_MOD_figure_render
#5  __fortplot_figure_core_io_MOD_render_figure_impl
...

Under gdb the fault is a len_trim on a null string:

#0  _gfortran_string_len_trim (len=..., s=0x0)
#2  __fortplot_subplot_rendering_MOD_render_subplots

Root cause: compute_tight_subplot_margins only supports the raster and PDF
backends, so for ASCII it returns ok=.false. and render_subplots takes
the non-tight margin branch. That branch reserves suptitle space with

if (allocated(state%suptitle) .and. len_trim(state%suptitle) > 0) then

which relies on short-circuit evaluation that Fortran does not guarantee (and
which the project rules explicitly forbid assuming). With no suptitle set,
len_trim dereferenced the unallocated string. Raster/PDF take the tight
path and skip this block, so they were unaffected.

Fix

Split the compound .and. into a nested if so len_trim never sees an
unallocated suptitle.

Verification

Build (serial to avoid the unrelated fpm parallel-build SIGABRT):

OMP_NUM_THREADS=1 fpm build

Failing before the fix (minimal 1x2 grid, no suptitle, save to .txt):

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
#3  __fortplot_subplot_rendering_MOD_render_subplots

Passing after the fix:

$ fpm test --target test_ascii_subplots_2019
 PASS: ASCII subplot save renders without crashing

Added test/ascii/test_ascii_subplots_2019.f90, which saves a subplot grid
(no suptitle) to ASCII and asserts the file is written and non-empty.

Related ASCII/subplot tests pass: test_ascii, test_subplots,
test_suptitle, test_utf8_text_preprocessing. Rendering gate
(make verify-artifacts) passes.