Today I learned that `rustfmt` is now a part of `rustup`. So I need
to actually look it up in PATH instead of using `cargo-fmt`.

I also learned that `mk` is the file that does the main work and is
shared by `cargo-new` and `cargo-init`, as expected.

Printing inside `mk` produces output as expected.

I think I found the correct place in the algorithm where rustfmt should
be called.

I'm not sure if using `std::process::Command` and converting errors into
warnings is the correct way here.

Also the warning should only be printed once. Unless
`config.shell().warn()` deduplicates warnings, right now there will be
one warning per source file.

So I should probably check that `rustfmt` exists only once and disable
formatting if it doesn't.

…ng-a-crate

There's now exactly one warning per new crate instead of one per
source file.

Also warnings are being logged now, not printed to console.

Test 1: init a library crate with a `rustfmt.toml` config file in it.
Expect the generated source files to be formatted according to the
config file.

Test 2: same as test 1, but with missing `rustfmt`. Expect `cargo init`
to ignore the absence of `rustfmt` and generate source files
with default formatting.

…ng-a-crate

The tests were failing with error:
"could not determine the current user, please set $USER"

Worst case is that the logs will have multiple errors caused by missing
rustfmt, so checking separately that rustfmt exists is unnecessary.

…ng-a-crate

…ng-a-crate

Generalized `clippy_is_available` and renamed as `command_is_available`.

No checks in `ignores_failure_to_format_source`, it's not supposed to
use `rustfmt` even if it's available