Fixed allocation-size-too-big error in H5MM.c

[bmribler]

A decoded length appeared to be corrupted and had a very large value.
This PR added a check to detect such potential data corruption.

The fuzzer file is in the cve_hdf5 repo.

Fixes GH-4431

[qkoziol]

Almost certainly, the cache is not the problem.

What kind of entry's "get_load_size" callback is being invoked to create the corrupted length?

[bmribler]

Almost certainly, the cache is not the problem.

What kind of entry's "get_load_size" callback is being invoked to create the corrupted length?

Thanks, Quincey. H5HL__cache_prefix_get_final_load_size/H5HL__hdr_deserialize:
heap.dblk_size = 18446744073709551392

[qkoziol]

OK, so it's in the local heap decode / get size callback. Is there any sanity checking you can do there?

[bmribler]

OK, so it's in the local heap decode / get size callback. Is there any sanity checking you can do there?

Oh, I think I can do a similar check in H5HL__cache_prefix_get_final_load_size. Right? Or maybe, H5HL__hdr_deserialize, if I have enough info...

[qkoziol]

Yes, adding a sensible range check there is good.