Implementing it in C makes it about 4x-6x faster

Microbenchmark:

```
 # bench.py

import time
import timeit
import asyncio

ITERS: int = 10**6
NANO: int = 10**9
NANO_PER_ITER: float = NANO / ITERS

async def main():
   # avoid load attr noise
   py_current_task = asyncio.tasks._py_current_task
   c_current_task = asyncio.tasks._c_current_task

   asyncio.current_task() # warmup
   py_current_task() # warmup
   c_current_task() # warmup

   print(
      "current_task: {}ns".format(timeit.timeit(py_current_task, number=ITERS, timer=time.process_time) * NANO_PER_ITER)
   )
   print(
      "current_task: {}ns".format(timeit.timeit(c_current_task, number=ITERS, timer=time.process_time) * NANO_PER_ITER)
   )

asyncio.run(main())
```

a few runs on MacBook Pro
2.4 GHz 8-Core Intel Core i9
64 GB 2667 MHz DDR4:

debug build:

```
~/work/pyexe/main-dbg ⌚ 9:57:34
$ ./python.exe bench.py
[py] current_task: 606.234ns
[c] current_task: 104.47699999999993ns

~/work/pyexe/main-dbg ⌚ 9:57:59
$ ./python.exe bench.py
[py] current_task: 631.856ns
[c] current_task: 110.22500000000002ns

~/work/pyexe/main-dbg ⌚ 9:58:08
$ ./python.exe bench.py
[py] current_task: 637.746ns
[c] current_task: 105.03899999999999ns

~/work/pyexe/main-dbg ⌚ 9:58:16
$ ./python.exe bench.py
[py] current_task: 621.3169999999999ns
[c] current_task: 103.01300000000002ns
```

opt build:

```
~/work/pyexe/main-opt ⌚ 10:33:17
$ ./python.exe bench.py
[py] current_task: 128.743ns
[c] current_task: 31.997999999999998ns

~/work/pyexe/main-opt ⌚ 10:33:24
$ ./python.exe bench.py
[py] current_task: 126.388ns
[c] current_task: 32.64599999999998ns

~/work/pyexe/main-opt ⌚ 10:33:26
$ ./python.exe bench.py
[py] current_task: 137.053ns
[c] current_task: 32.066999999999986ns

~/work/pyexe/main-opt ⌚ 10:33:28
$ ./python.exe bench.py
[py] current_task: 131.17800000000003ns
[c] current_task: 32.06600000000001ns
```

….current_task for ~40% speedup

… current_task