RUST-2032 track task spawn location for tokio instrumentation

[hds]

The tracing instrumentation present in tokio (which feeds into
tokio-console) tracks the spawning location of each task. This can be
useful in understanding what each task is doing, especially if the tasks
aren't named (which is a tokio_unstable feature so not usually present
in published crates).

The location tracking isn't as useful when tasks are spawned from some
central location like the mongodb crate does, because all tasks have
the same spawn location. However, since the instrumentation uses the
panic:Location to determine the spawn location, the #[track_caller]
attribute macro can be used to ensure that the actual location is picked
up instead of the common one.

A side effect of using the #[track_caller] attribute is that the
single line output in the case of a panic (when RUST_BACKTRACE isn't
set) will point to the callsite in the code, and not the common site -
although this is often desirable anyway.

This change adds #[track_caller] to the pub(crate) function
crate::runtime::spawn as well as the method AsyncJoinHandle::spawn.

[hds]

A couple of notes on the MR. This came about after receiving a question about the behaviour of an application which was using the mongodb crate and had this screenshot from tokio-console:

All those tasks would have a more useful value in the Location column with this patch.

No tests are included because testing #[track_caller] is a little bit involved - and testing it with the tokio instrumentation enabled even more so. However, I'm happy to add tests by forcing a panic and recording the panic location, they would look something like the ones here:

https://github.com/tokio-rs/tokio/blob/91169992b2ed0cf8844dbaa0b4024f9db588a629/tokio/tests/rt_panic.rs

with the following helper function:

https://github.com/tokio-rs/tokio/blob/91169992b2ed0cf8844dbaa0b4024f9db588a629/tokio/tests/support/panic.rs

[abr-egn]

This looks like a very nice enhancement, thank you! I've kicked off the test run and if there aren't any known failures I'll merge this in :)