Improve writer and verifier performance by using multiple threads

[ifd3f]

Is your feature request related to a problem? Please describe.

Both of these are single-threaded operations. This is why decompressed writing is super slow. Yeah, I know, amateur hour.

Describe the solution you'd like

We will split writer and verifier into multiple reusable threads that communicate to each other likely via mpsc::channel<Bytes>() or mpsc::channel<Arc<[u8]>>():

• File reading thread
• Decompressing thread
• Fanout thread (take a stream of bytes and spread it out to N sinks, to support Multi-disk UI  #64)
• Disk writer thread
• Disk reader thread
• Byte-by-byte verifier thread
• (eventually)
  • Hasher thread (to support Allow burning a file straight from stdin #137)
  • Not a thread, a whole separate process for decompression (Use shell executable versions of compression and hash algs instead of rust versions #158)

After this, the logic for spawning writer and verifier tasks will just be wiring.

This can be done simultaneously with #234.

Alternatives

Why mpsc::channel<Bytes>() or mpsc::channel<Arc<[u8]>>() instead of some exotic datastructure? Because it's easy to work with. Plus, disk is probably going to be the limiting factor, not atomic reference counts.

I would like to optimize my pipelines later in Exotic Ways. However, we shouldn't do this until we have good benchmarking (see #238)

[ifd3f]

#262 has laid the groundwork to get this done.