Fix off-by-one spans in MIR borrowck errors

Fixes #46885.

r? @nikomatsakis

Libtest json output

A revisit to my [last PR](https://github.com/rust-lang/rust/pull/45923).

Events are now more atomic, printed in a flat hierarchy.

For the normal test output:
```
running 1 test
test f ... FAILED

failures:

---- f stdout ----
	thread 'f' panicked at 'assertion failed: `(left == right)`
  left: `3`,
 right: `4`', f.rs:3:1
note: Run with `RUST_BACKTRACE=1` for a backtrace.

failures:
    f

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out
```

The JSON equivalent is:
```
{ "type": "suite", "event": "started", "test_count": "1" }
{ "type": "test", "event": "started", "name": "f" }
{ "type": "test", "event": "failed", "name": "f" }
{ "type": "suite", "event": "failed", "passed": 0, "failed": 1, "allowed_fail": 0, "ignored": 0,  "measured": 0, "filtered_out": "0" }
{ "type": "test_output", "name": "f", "output": "thread 'f' panicked at 'assertion failed: `(left == right)`
  left: `3`,
 right: `4`', f.rs:3:1
note: Run with `RUST_BACKTRACE=1` for a backtrace.
" }
```

For E0277 on `for` loops, point at the "head" expression

When E0277's span points at a `for` loop, the actual issue is in the
element being iterated. Instead of pointing at the entire loop, point
only at the first line (when possible) so that the span ends in the
element for which E0277 was triggered.

Match libunwind's EABI selection with libpanic_unwind

Currently, the `libunwind` crate will only select the ARM EABI if it is compiling for ARM/Linux or Android targets. `libpanic_unwind`, however, will choose the ARM EABI if the target arch is ARM and the OS is not iOS. This means that if one tries to enable unwinding for a non-standard ARM target (such as implementing a custom stdlib via Xargo, for example), then the two crates can potentially disagree about which EABI is being targeted.

This PR makes `libunwind` use the [same logic](https://github.com/rust-lang/rust/blob/master/src/libpanic_unwind/gcc.rs#L139-L146) as `libpanic_unwind` when choosing the EABI.

I noticed there are a few comments about certain functions only differing on Android or ARM/Linux, but I *think* that those differences apply to the ARM EABI in general. Let me know if I'm wrong about that.

Modify the spans used for `for`-loop expression expansion, instead of
creating a new span during error creation.

libtest: Split HumanFormatter into {Pretty,Terse}

libtest: Fixed padding of benchmarks when not benchmarking

libtest: Fixed benchmarks' names not showing in terse-mode

libtest: Formatting

libtest: Remove usage of jq

libtest: Fixed UI tests

- Now comparing to the right file.
- A python script checks for validity of JSON documents

libtest: Whoops

libtest: Added the -Z option for unstable options

libtest: Json format now outputs failed tests' stdouts.

libtest: Json format now outputs failed tests' stdouts.

libtest: Json formatter now spews individiual events, not as an array

libtest: JSON fixes

libtest: Better JSON escaping

libtest: Test start event is printed on time

Rollup of 19 pull requests

- Successful merges: #47415, #47437, #47439, #47453, #47460, #47502, #47529, #47600, #47607, #47618, #47626, #47656, #47668, #47696, #47701, #47705, #47710, #47711, #47719
- Failed merges: #47455, #47521

Print inlined functions on Windows

Split from https://github.com/rust-lang/rust/pull/45637

r? @alexcrichton

Fixes #47364