cksum: incomplete hex checksum is treated as mismatch instead of improper format

[BenWiederhake]

When checking a cksum-file, hexstrings that have the wrong length are reported as mismatching errors instead of simply broken formatting. (Arguably, they should be treated as errors for safety, but I guess it's better to stick with GNU behavior here.)

$ cat sums
BLAKE2b (LICENSE) = e45898002c601e57504d0a7d7cb7ea5a1bc58d2f9a94fde09f370045ba2b3d57cbd26c3b0df1b4023b08c8b101303bb1b1b7c3363b38085194c1067c487a3052
BLAKE2b (build.rs) = ab3ec55005267c4d6333c585a313d424dc3fffc4f5b29aa335724d444bf6a9b26234b9db506e92b48dc76800248fffe618889d45a7b3b5228baca2ff7e2db3e
BLAKE2b (README.md) = c0a230cfdbbb686f91071119c915a19df28cb2e4fc575f19d5c6b81bf4e3ca2e83e6df49f14d8b89ece14eb4090be435a7af6fed2c981774e076caa8fe62b0
$ cksum -c sums
LICENSE: OK
cksum: WARNING: 2 lines are improperly formatted
$ cargo run -q cksum -c sums
LICENSE: OK
build.rs: FAILED
README.md: FAILED
cksum: WARNING: 2 computed checksums did NOT match
[$? = 1]

This isn't specific to blake2b:

$ cat sums
SHA256 (LICENSE) = 29953405eaa3dcc41c37d1621d55b6a47eee93e05613e439e73295029740b10c
SHA256 (build.rs) = 299aa98fed9979a78d4022c9993ac19a5c647167b4e85ba7af2240d91ea759d
SHA256 (README.md) = 4d9b4c7b50b976ec5b0b2dce5a767d2da4b63eb68592017d475c4e11799a33
$ cksum -c sums
LICENSE: OK
cksum: WARNING: 2 lines are improperly formatted
$ cargo run -q cksum -c sums
LICENSE: OK
build.rs: FAILED
README.md: FAILED
cksum: WARNING: 2 computed checksums did NOT match
[$? = 1]

Found while reviewing #6499, although it's not the cause of this bug. (The expected_checksum.len() / 2 expression looks suspicious, though.)

[cakebaker]

Reopening this ticket as it was closed by accident due to merging #6868 .

[sylvestre]

i don't think it is covered by upstream tests btw as we are at 100 % coverage here:
https://uutils.github.io/coreutils/docs/test_coverage.html

[RenjiSann]

i don't think it is covered by upstream tests btw as we are at 100 % coverage here: https://uutils.github.io/coreutils/docs/test_coverage.html

There are indeed several cases that are not tested correctly by the GNU testsuite, namely all the issues we found, even though we pass all the GNU tests