Split core functionality into a serenity-core subcrate

[mkrasnitski]

This PR splits serenity into two crates, serenity and serenity-core.

• The core library contains the builder, cache, http, model, and utils modules.
• The serenity crate depends on the core and re-exports everything in it, plus it also contains the gateway, collector, framework, and interactions_endpoint modules. It also re-exposes all of the core's feature flags, though I'm not sure if that's the right approach.

Further splitting may be possible in the future but would require significant refactoring to remove circular dependencies between modules. For error propagation, a bunch of variants on serenity::Error got moved into the core, and are propagated using Error::Core which wraps a serenity_core::Error. Though, I wonder if it's better to keep the variants and convert 1-1 from a serenity_core::Error so as to keep things more flat.

[mkrasnitski]

Ran some compile-time benchmarks on my machine using hyperfine and came up with some numbers. Before this PR:

Benchmark 1: cargo check
  Time (mean ± σ):     14.868 s ±  0.049 s    [User: 60.332 s, System: 9.855 s]
  Range (min … max):   14.760 s … 14.931 s    10 runs

Benchmark 2: cargo build
  Time (mean ± σ):     19.035 s ±  0.059 s    [User: 83.263 s, System: 11.827 s]
  Range (min … max):   18.945 s … 19.098 s    10 runs

Benchmark 3: cargo build --release
  Time (mean ± σ):     24.560 s ±  0.291 s    [User: 195.193 s, System: 12.501 s]
  Range (min … max):   24.159 s … 25.139 s    10 runs

After this PR:

Benchmark 1: cargo check
  Time (mean ± σ):     14.478 s ±  0.054 s    [User: 59.951 s, System: 9.903 s]
  Range (min … max):   14.377 s … 14.572 s    10 runs

Benchmark 2: cargo build
  Time (mean ± σ):     17.117 s ±  0.036 s    [User: 82.604 s, System: 12.095 s]
  Range (min … max):   17.079 s … 17.185 s    10 runs

Benchmark 3: cargo build --release
  Time (mean ± σ):     22.155 s ±  0.143 s    [User: 199.097 s, System: 12.608 s]
  Range (min … max):   21.930 s … 22.431 s    10 runs

If we compare mean execution times, that gives a 2.6% improvement for cargo check, 10.1% improvement for cargo build, and 9.8% improvement for cargo build --release.

[mkrasnitski]

This PR is ready for review now

[GnomedDev]

Looks good, and a good start to splitting apart serenity.

I would be interested in seeing what the difference in cargo build --timings looks like, specifically with a poise or songbird based bot to see if they can depend only on core and kick off their build earlier.

I haven't taken a deep look over this, but spotted this comment immediately.