Group `target_module`, `path` and `file` arguments

[EFanZh]

This is the first step of bring optimizations from #569 into the log crate.

This PR contains two commits:

• The first commit moves all private APIs into a single __private_api module so that it is easier to implement more elaborated optimizations.
• The second commit splits &(target, module_path, file, line) argument into two arguments: &(target, module_path, file) and line. The idea is that a module could have more than one log calls, so most likely, these log calls would have the same (target, module_path, file) triple, group them together will allow multiple log calls to reuse a same (target, module_path, file) instance, see https://godbolt.org/z/z7jW67Ece for comparison.

For one of my internal projects, this PR will reduce the binary size by about 2%.

I also tested this PR on cargo-binstall, this is the test result:

opt-level	lto	codegen-units	original size	optimized size	diff
3	off	1	14211752	14199464	-12288
3	off	16	16739048	16739048	0
3	thin	1	15170120	15166024	-4096
3	thin	16	17566376	17566376	0
3	fat	1	13859336	13859336	0
3	fat	16	15231496	15231496	0
"z"	off	1	11451048	11438760	-12288
"z"	off	16	13531848	13511368	-20480
"z"	thin	1	12307048	12298856	-8192
"z"	thin	16	13699720	13695624	-4096
"z"	fat	1	9951752	9935368	-16384
"z"	fat	16	10476040	10459656	-16384

[EFanZh]

OK, I have tested the performance using this code:

use criterion::Criterion;

fn log_string_literal(c: &mut Criterion) {
    let mut benchmark_group = c.benchmark_group("log string literal");

    benchmark_group.bench_function("original", |b| b.iter(|| log::info!("abc")));
    benchmark_group.bench_function("optimized", |b| b.iter(|| log_optimized::info!("abc")));

    benchmark_group.finish();
}

criterion::criterion_group!(benches, log_string_literal);

criterion::criterion_main!(benches);

with the following dependencies in Cargo.toml:

criterion = "*"
log = "*"
log_optimized = { git = "ssh://git@github.com/EFanZh/log.git", branch = "group-target-module-path-and-file", package = "log" }

On average, the original one takes 246.84 ps per iteration, and the optimized one takes 245.54 ps per iteration, I consider this difference insignificant.

[EFanZh]

Hi @KodrAus, I’d really appreciate it if you could take a look at this PR. I intend to do some binary size optimizations with ideas from #569. I plan to test and implement each idea with a separate PR, so if this PR gets accepted, I can move on to the next PR. Will these sort of optimizations be accepted?