Replace all libz-sys code with flate2

[cormacrelf]

Much less prone to errors; reads directly into pdf_stream Vec<u8>.

Relevant: #171

[cormacrelf]

It's not the memLevel setting.

There appear to be three things responsible for the diffs:

• The allocators! flate2 uses its own allocators, based on std::alloc. Swapping them in within a Rust version of compress2 gets different compression results.
• flate2 also puts zlib headers in there by providing window_bits = -15 instead of 15.
• flate2 also sets the flush value differently, to Z_NO_FLUSH where possible. This lets the algorithm control when to flush.

In any case, the PDFs all seem to open just fine. Zlib will always be able to decompress its own successful output.

[cormacrelf]

If this is inaccurate, I'm not sure there's really a backwards-compatibility point in keeping the zlib linking. I think the default backend runs about as quickly as zlib.

[cormacrelf]

With zlib, 273 diffs: https://tt.ente.ninja/#/compare/c188c51eadc8dfcc46b77819b7b8ac7d374e582b/5a5e070ef6bbdeb6e06d1956233d31e402a158cd

[cormacrelf]

With default, every file is different.