gh-87909: avoid a 0 bytes copy call to sendfile()

[gpshead]

That triggers an EAGAIN error on a minority of kernels and filesystems.

Based on a patch suggestion by pmrv (see the linked issue).

• Issue: BlockingIOError: [Errno 11] Resource temporarily unavailable: on GPFS. #87909

[gpshead]

I turned the old patch on the issue into this PR... it's clearly not right based on existing tests:

test_exception_on_second_call (test.test_shutil.TestZeroCopySendfile.test_exception_on_second_call) ... FAIL
test_small_chunks (test.test_shutil.TestZeroCopySendfile.test_small_chunks) ... FAIL
======================================================================
FAIL: test_exception_on_second_call (test.test_shutil.TestZeroCopySendfile.test_exception_on_second_call)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/runner/work/cpython/cpython-ro-srcdir/Lib/test/test_shutil.py", line 2565, in test_exception_on_second_call
with self.assertRaises(OSError) as cm:
AssertionError: OSError not raised
======================================================================
FAIL: test_small_chunks (test.test_shutil.TestZeroCopySendfile.test_small_chunks)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/runner/work/cpython/cpython-ro-srcdir/Lib/test/test_shutil.py", line 2591, in test_small_chunks
self.assertEqual(read_file(TESTFN2, binary=True), self.FILEDATA)
AssertionError: b'Qcj[83[886](https://github.com/python/cpython/actions/runs/3977715463/jobs/6819054400#step:17:887)00 chars]vKQMs' != b'Qcj[8388600 chars]vKQMsQcjlcBkZjwawMTZtHYhdRhukiOdKiAFKthzsKfEAa[2097107 chars]KQMs'
----------------------------------------------------------------------
Ran 2 tests in 1.977s
FAILED (failures=2)
test test_shutil failed
1 test failed again:
test_shutil

[hugovk]

This old patch didn't work out, shall we close it?