fix: compression priority

[inetol]

Before this PR, fasthttp would never use zstd if clients sent their Accept-Encoding header including gzip because zstd was the last option and would only be used if zstd was the only encoding available in Accept-Encoding header. This is really an issue when serving precompressed sidecars of zstd and gzip where the server preferred the latter.

The following order has been established; first brotli, then zstd, and finally gzip. It takes into account whether CompressBrotli and the new CompressZstd option are enabled to select which compression to use.

It also fixes the missing Vary header on compressed assets.

There is also an issue that has been revealed with the rework of these tests that should be addressed in a separate PR

[erikdubbelboer]

Nice, I like the change. It makes sense to try better algorithms before falling back to gzip.

Can you have a look are the linting issues and failing tests?

[inetol]

Everything should be ready now, the failed tests that appear is a specific Windows flaky thing?

[erikdubbelboer]

Thanks!