Issue-118, Auto compressor selection fix

[worryg0d]

This PR provides a fix for a bug described in #118.

CompressorFrame::compress_best picks the best compressor by comparing its space usage, but it doesn't consider the max_error allowed. This PR fixes this problem by filtering the results by max_error before picking a compressor that compresses the best.

Closes #118

[worryg0d]

Test fails because CompressorFrame::compress_best filters all compressors out because some returned inf and NaN values as computed errors.

I'm not sure how to deal with this one. Should it return some default value? Or is panic not a problem here?

[worryg0d]

@cjrolo Hello!

Could you please look at this and #117 PRs?

[cjrolo]

Should I do this first or #117 ?

[worryg0d]

Should I do this first or #117 ?

I think this one is more prioritised, because it actually should allow #117 to pass tests

[cjrolo]

Ok, I think I've found the issue! Look into the code and if you agree with it, merge it!

[rukai]

I pushed a commit that maintains the same logic as carlos' commit but reduces the allocations.

Previously we were cloning the results Vec<u8> stored in the iterator, one for each compression.
This results in a new allocation for each compression type (currently 2 but that could change in the future) and also a potentially large memcpy for each.

With this commit we just allocate once for the vec to hold each compression.

I also bumped to the latest rust version to make use of the new lint ignore reason field.