Add support for custom demanglers

[DaniPopes]

Adds support for custom demangler functions with the signature fn(&str, &mut Buffer) -> std::fmt::Result where Buffer is an opaque type that only exposes an implementation of std::fmt::Write.

The default implementation is moved to a standalone function, and the buffer is now cleared before any demangling is done as tracy guarantees:

When a call to ___tracy_demangle is made, previous contents of the string memory
do not need to be preserved.

Follow-up to #102.

[nagisa]

This won't work right in MT contexts, will it? Don't we need at least a buffer for each thread?

[DaniPopes]

Don't think it has to be MT-safe, the cpp implementation uses a global malloc'd pointer that's reused like we do here and it's worked fine

[nagisa]

I would also pub use crate::demangle::register_demangler (see a similar thing done with frame above.)

[DaniPopes]

It's already exported through macro_export, do you want to do this regardless?

[nagisa]

Yes. Among other reasons it makes use crate::register_demangler work within the crate without surprises...

[DaniPopes]

error[E0432]: unresolved import `crate::demangle::register_demangler`
  --> tracy-client/src/lib.rs:31:9
   |
31 | pub use crate::demangle::register_demangler;
   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ no `register_demangler` in `demangle`
   |
   = note: this could be because a macro annotated with `#[macro_export]` will be exported at the root of the crate instead of the module where it is defined

[nagisa]

Merged as 24a868a after a rebase.