fix(memcached): let clear_path raise connection errors

[allen0099]

Closes #177.

Problem

MemcachedBackend.clear_path() wrapped its delete in except Exception: return 0. A connection failure was therefore reported as "nothing to clear", while delete() and every other method let the error through.

Change

Drop the try/except. clear_path() now reports errors the same way delete() does.

Note on pymemcache's own error handling

HashClient (ignore_exc=False, default retry_attempts=2, retry_timeout=1, dead_timeout=60) behaves like this:

• It raises the OSError on the first failed call.
• For the next second, every call on that server returns the command's default (None/False) without raising.
• After the retries, it raises MemcacheError("All servers seem to be down right now").

This PR only removes the extra layer in clear_path(), which also hid the first error and the "all servers down" one. The silent defaults during the retry window affect every method, and are tracked in #197.

Tests

• New test_memcached_clear_path_raises_when_the_server_is_unreachable points the backend at a closed port and expects ConnectionRefusedError. It needs no live server.
• test_memcached_clear_path_exception asserted that the error was swallowed. It is now ..._propagates_client_errors, and expects the error.
• Mutation check: with the except restored, exactly these two tests fail.
• Full suite against live Redis and Memcached (CACHEX_REQUIRE_LIVE_SERVERS=1): 849 passed, 100% coverage.