Retire map_exceptions

[akx]

Summary

try: except: blocks are free on modern Pythons when exceptions aren't raised.

Entering a @contextmanager'd with block is much less free (see e.g. this Django discussion about contextlib.suppress), and there are hot paths (e.g. reading from a sync socket) where it's worth avoiding that overhead.

For consistency, this retires the use of map_exceptions everywhere.

Checklist

• [x] I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• [x] I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
  • No test changes needed.
• [x] I've updated the documentation accordingly.
  • No changes needed; internal change.

Performance impact

Using pytest-benchmark, and Caddy locally serving a 10-megabyte file,

from httpcore import ConnectionPool


def test_bench(benchmark):
    with ConnectionPool() as pool:
        def b():
            resp = pool.request(
                method="GET",
                url="http://127.0.0.1:8080/bigfile2",
            )
            assert len(resp.content) == 10 * 1024 * 1024
        benchmark(b)

is shown to be a bit faster:

Name (time in ms)               Mean                 OPS
------------------------------------------------------------------
test_bench (0002_2655597)     3.8573 (1.0)      259.2509 (1.0)
test_bench (0001_5974b03)     4.2206 (1.09)     236.9331 (0.91)
------------------------------------------------------------------

[akx]

The new coverage misses are a bit :shrug:, because evidently those cases weren't tested previously either, they were just conveniently covered by the exception mapping 😄