The handling of N-to-M array compression was broken with the addition
of the partial buffer option, which broke the bounds check that was
being used.
Note that this also makes the partial buffer option more consistent;
if, for example, you have
```
record(compress, foo) {
field(ALG, "N to 1 Average")
field(INP, "bar NPP")
field(NSAM, 2)
field(N, 2)
field(PBUF, YES)
}
```
(with `bar` having, e.g. length 3), then this will now behave as
expected on both of the samples.
Note that it is not really a circular buffer in this case, but
a full reset of the buffer to the beginning. This matches the
documentation, but it seems valuable to add an explicit test for
this case.
In some cases the license-identification header was missing,
so I added that as well. Replaced the remaining headers that
specifically identified "Versions 3.13.7 and higher".
Makefiles and the build system were deliberately excluded.
