Skip manifest loading if there are no components/targets to check

[Kobzol]

While examining the manifest loading issues in #2626, I noticed that for the common case (running just rustc), the list of components and targets to check is actually empty. In that case, we parse the manifest for no reason, costing us ~20-30ms on every rustup proxy invocation.

This makes the common case of running rustc almost as fast as when the toolchain is specified explicitly using e.g. +nightly, where the components/targets weren't checked before.

This does not help the situation where there is a rust-toolchain.toml file with custom components and targets, these will be checked as before.

Before:

$ cargo build --release && cp target/release/rustup-init rustc
$ hyperfine "./rustc +nightly --version" "./rustc --version"

Benchmark 1: ./rustc +nightly --version
  Time (mean ± σ):      17.3 ms ±   0.4 ms    [User: 5.2 ms, System: 13.3 ms]
  Range (min … max):    16.6 ms …  19.0 ms    173 runs
 
Benchmark 2: ./rustc --version
  Time (mean ± σ):      44.8 ms ±   0.6 ms    [User: 25.5 ms, System: 20.6 ms]
  Range (min … max):    43.7 ms …  46.4 ms    65 runs
 
Summary
  ./rustc +nightly --version ran
    2.59 ± 0.07 times faster than ./rustc --version

After:

$ cargo build --release && cp target/release/rustup-init rustc
$ hyperfine "./rustc +nightly --version" "./rustc --version"

Benchmark 1: ./rustc +nightly --version
  Time (mean ± σ):      17.3 ms ±   0.4 ms    [User: 5.3 ms, System: 13.1 ms]
  Range (min … max):    16.4 ms …  19.2 ms    170 runs
 
Benchmark 2: ./rustc --version
  Time (mean ± σ):      17.4 ms ±   0.5 ms    [User: 5.3 ms, System: 13.3 ms]
  Range (min … max):    16.5 ms …  18.8 ms    165 runs
 
Summary
  ./rustc +nightly --version ran
    1.01 ± 0.04 times faster than ./rustc --version

Alternative to #4344.
Related issue: #2626

[djc]

Awesome find!