Fixes adjustment of max_width within macros

[csmulhern]

Fixes #5404.

Minimal example:

macro_rules! foo {
    () => {
        println!("Hello, world!");
    };
}

Without this change:

> cargo run --bin rustfmt -- --config-path=/dev/null --config=max_width=100 --config=fn_call_width=100 ~/Desktop/example.rs
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.02s
     Running `target/debug/rustfmt --config-path=/dev/null --config=max_width=100 --config=fn_call_width=100 /Users/cameron/Desktop/example.rs`
`fn_call_width` cannot have a value that exceeds `max_width`. `fn_call_width` will be set to the same value as `max_width`

With this change:

> cargo run --bin rustfmt -- --config-path=/dev/null --config=max_width=100 --config=fn_call_width=100 ~/Desktop/example.rs
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.03s
     Running `target/debug/rustfmt --config-path=/dev/null --config=max_width=100 --config=fn_call_width=100 /Users/cameron/Desktop/example.rs`

[csmulhern]

@ytmimi, could you take a look? Happy to make changes if necessary.

[csmulhern]

@Manishearth I see you merging some recent PRs. Do you know who can take a look at this?

[Manishearth]

@csmulhern So we're trying to help spread out the maintainer burden, but there are a lot of open PRs right now and we're still triaging them. Open PRs which change formatted code are tricky because it needs to be determined if it's a rbeaking change or not.

We'll get to them eventually, but no promises.

It will help if you show a before and after of the formatted code.

[Manishearth]

It seems like from the PR description that this doesn't actually impact formatting, this makes certain previously-unintentionally-invalid values of a config start working again.

[csmulhern]

Yes, this (mostly*) shouldn't impact formatting.

Today, we have a bunch of properties that control breakpoints for how long certain lines can be. This is primarily max_width, but also fn_call_width, attr_fn_like_width, struct_lit_width, struct_variant_width, array_width, chain_width, single_line_if_else_max_width. By default, these are all derived from max_width, due to use_small_heuristics defaulting to true.

In order to support formatting in macros, it looks like rustfmt just trims down max_width by the indentation of the macro arm body, and then runs formatting of the body with that adjusted config. The problem is, those other max_width style properties are not adjusted. So then when those values are read inside the formatting for the macro body, they can be larger than max_width, which triggers this error, and the values are clamped down to max_width. Note that if these values are unset, there's no issue, because it will just default to the heuristic value.

This change makes the reduction based on the indent symmetric across all of the max_width style properties, and adds a new method to ensure this is done universally.

* The edge case where this would affect formatting would be if e.g. max_width was set to 80, and fn_call_width was set to 76, and the macro body indent was 8. In today's world, fn_call_width in the macro body would be larger than the max_width within the macro body (72), and would get clamped down to 72. With this fix, both would be reduced by 8 (max_width 80 -> 72, fn_call_width 76 -> 68).