Fix byte offsets of row oriented array components and out-of-record reads in get_channel_raw_bytes

[victorjoh]

Fixes #1318. Related: #1264, #1316.

• ChannelArrayBlock._factors: for row-oriented layout the last dimension now varies fastest, so f(d) = f(d+1)·dim_size(d+1). Column-oriented (FLAG_CA_INVERSE_LAYOUT) results are unchanged.
• get_channel_raw_bytes / get_channel_raw_bytes_C: clamp the number of bytes read to the part of the channel inside the record, and zero-fill the rest. Before this, a channel starting at or after the end of the record wrote past the output buffer.

Tests:

• test_mdf4.py::test_row_oriented_array_components: writes a 7 × 6 matrix and checks that all element channels are inside the record and have the right values.
• test_cutils.py: get_channel_raw_bytes with channels inside, partly outside and fully outside the record. The last case only fails reliably under PYTHONMALLOC=debug.

Both new tests fail on development and pass with this change.

[danielhrisca]

thanks @victorjoh

[derdabi]

Thanks for merging this. Is there a planned release that will include the fix, e.g. the next release from development? Since there are other fixes there as well, we'd like to know roughly when to expect it. If that's still some way off, a backport to asammdf_8.8.x (8.8.28) would also help us, as we're on 8.8.27 and affected by the wrong byte offsets of row-oriented array components.