Avoid boxing large futures

[james7132]

Fixes #83. Fixes #66. Partially fixes #78.

This PR gets rid of the extra boxing in spawn_unchecked that results in extra monomorphization and overhead of allocating and moving into a Pin<Box<Fut>>. The boxing causes an extra copy onto the stack before moving onto the heap. This negates only one of the two stack copies made on lower optimization builds, and requires an extra indirection on every access, so we get rid of it.

Instead, this PR turns spawn_unchecked into a macro to avoid creating extra stack frames, effectively forcibly inlining the code regardless of what optimization level. This PR also opts to replace the abort_on_panic implementation with a manual one to avoid the extra stack copy there.

[james7132]

I'm not sure where the CI failures are coming from. I don't think they're related to the changes here, and they show up in #87 as well. @taiki-e FYI.

[taiki-e]

LGTM

[james7132]

@taiki-e any thoughts on the CI failures? I'm not sure where the leaks are coming from.

[taiki-e]

All errors come from doctest, so I think it's likely because doctest-xcompile has stabilized in 1.89 and doctests are now tested even when runner is set. (So it's not due to this PR.)

failures:
    src/lib.rs - (line 25)
    src/runnable.rs - runnable::Runnable<M>::schedule (line 727)
    src/runnable.rs - runnable::Runnable<M>::waker (line 790)

All "definitely lost" indicate a leak of allocation from flume::unbounded that is captured in the closure, so this is probably due to the task being scheduled but not polled.

==6502== 248 (152 direct, 96 indirect) bytes in 1 blocks are definitely lost in loss record 3 of 4
==6502==    at 0x48C480F: malloc (vg_replace_malloc.c:446)
==6502==    by 0x402D07D: alloc::alloc::Global::alloc_impl (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x402CF59: alloc::alloc::exchange_malloc (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x402CD51: alloc::sync::Arc<T>::new (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x402F523: flume::unbounded (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x402FCDB: rust_out::main::_doctest_main_src_runnable_rs_790_0 (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x402FCA5: rust_out::main (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x402A512: core::ops::function::FnOnce::call_once (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x4029E95: std::sys::backtrace::__rust_begin_short_backtrace (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x4029E78: std::rt::lang_start::{{closure}} (in /tmp/rustdoctest3ICG2R/rust_out)
==6502==    by 0x404249F: call_once<(), (dyn core::ops::function::Fn<(), Output=i32> + core::marker::Sync + core::panic::unwind_safe::RefUnwindSafe)> (function.rs:290)
==6502==    by 0x404249F: do_call<&(dyn core::ops::function::Fn<(), Output=i32> + core::marker::Sync + core::panic::unwind_safe::RefUnwindSafe), i32> (panicking.rs:590)
==6502==    by 0x404249F: catch_unwind<i32, &(dyn core::ops::function::Fn<(), Output=i32> + core::marker::Sync + core::panic::unwind_safe::RefUnwindSafe)> (panicking.rs:553)
==6502==    by 0x404249F: catch_unwind<&(dyn core::ops::function::Fn<(), Output=i32> + core::marker::Sync + core::panic::unwind_safe::RefUnwindSafe), i32> (panic.rs:359)
==6502==    by 0x404249F: {closure#0} (rt.rs:175)
==6502==    by 0x404249F: do_call<std::rt::lang_start_internal::{closure_env#0}, isize> (panicking.rs:590)
==6502==    by 0x404249F: catch_unwind<isize, std::rt::lang_start_internal::{closure_env#0}> (panicking.rs:553)
==6502==    by 0x404249F: catch_unwind<std::rt::lang_start_internal::{closure_env#0}, isize> (panic.rs:359)
==6502==    by 0x404249F: std::rt::lang_start_internal (rt.rs:171)
==6502==    by 0x4029E60: std::rt::lang_start (in /tmp/rustdoctest3ICG2R/rust_out)