Fix ndarray dimension signedness, fix ndarray length overflow (2); add 32bit unit test

[falbrechtskirchinger]

Replaces #3521.

Closes #3519.

[falbrechtskirchinger]

@nlohmann Hm. The automatic detection part is failing spectacularly. Also, the CI image is missing some bits to make this work.
That means there's no other choice than to exclude the 32bit lines from coverage for now. I'll add a flag for enabling the 32bit test instead.

[falbrechtskirchinger]

I've verified that -DJSON_32bitTest=ON does in fact produce a working 32bit executable:

$ cmake ~/projects/json/issue3519-trait -DCMAKE_CXX_COMPILER=g++-12.1.1 -DJSON_MultipleHeaders=ON -DJSON_32bitTest=ON
$ make test-32bit_cpp11
$ file tests/test-32bit_cpp11 
tests/test-32bit_cpp11: ELF 32-bit LSB pie executable, Intel 80386, version 1 (GNU/Linux), dynamically linked, interpreter /lib/ld-linux.so.2, for GNU/Linux 3.2.0, with debug_info, not stripped
$ ./tests/test-32bit_cpp11
[doctest] doctest version is "2.4.6"
[doctest] run with "--help" for options
===============================================================================
[doctest] test cases:  6 |  6 passed | 0 failed | 0 skipped
[doctest] assertions: 23 | 23 passed | 0 failed |
[doctest] Status: SUCCESS!

[coveralls]

Coverage remained the same at 100.0% when pulling fc6ac6a on falbrechtskirchinger:issue3519-trait into b6d00d1 on nlohmann:develop.

[nlohmann]

Looks good to me.

[nlohmann]

Thanks! Let's see whether the fuzzer is satisfied now :)