Implement response heading parsing in rust (#54)

## Motivation / Description
There was nothing else to optimize in python, but Rust can parse
the string and build a ResponseFlags object in 22-50ns depending
on the number of flags.

While the interface between rust and python will never reach ns
performance, it helps indeed finding the header in the
response buffer.

Initial (before all optimizations):
multithreaded: Overall: 110779.55 RPS / 9.03 us/req
singlethreaded: Overall: 111545.63 RPS / 8.96 us/req

Python optimized:
multithreaded: Overall: 193340.40 RPS / 5.17 us/req
singlethreaded: Overall: 193036.56 RPS / 5.18 us/req

Using rust for header parsing:
multithreaded: Overall: 245898.34 RPS / 4.07 us/req
singlethreaded: Overall: 246165.19 RPS / 4.06 us/req

noxfile.py

@@ -6,7 +6,7 @@
 nox.options.sessions = "lint", "types", "tests"
 locations = "src", "tests", "noxfile.py"
 DEFAULT_VERSION = "3.8"
-DEFAULT_BENCHMARK_VERSION = "3.11"
+DEFAULT_BENCHMARK_VERSIONS = ["3.11"]
 VERSIONS = ["3.8", "3.11"]
 
 
@@ -57,7 +57,7 @@ def tests(session: Session) -> None:
     session.run("pytest", *args, env={"PYTHONHASHSEED": "0"})
 
 
-@session(python=DEFAULT_BENCHMARK_VERSION)
+@session(python=DEFAULT_BENCHMARK_VERSIONS)
 def benchmark(session: Session) -> None:
     """Run the benchmark suite."""
     args = session.posargs

pyproject.toml

@@ -14,6 +14,7 @@ packages = [{include = "meta_memcache", from="src"}]
 python = "^3.8"
 uhashring = "^2.1"
 marisa-trie = "^1.0.0"
+meta-memcache-socket = "^0.1.0"
 
 [tool.poetry.group.extras.dependencies]
 prometheus-client = "^0.17.1"

src/meta_memcache/__init__.py

@@ -40,6 +40,7 @@
     Miss,
     NotStored,
     ServerVersion,
+    ResponseFlags,
     SetMode,
     Success,
     TokenFlag,

src/meta_memcache/commands/high_level_commands.py

@@ -287,23 +287,23 @@ def get_or_lease_cas(
 
             if isinstance(result, Value):
                 # It is a hit.
-                if result.win:
+                if result.flags.win:
                     # Win flag present, meaning we got the lease to
                     # recache/cache the item. We need to mimic a miss.
-                    return None, result.cas_token
-                if result.size == 0 and result.win is False:
+                    return None, result.flags.cas_token
+                if result.size == 0 and result.flags.win is False:
                     # The value is empty, this is a miss lease,
                     # and we lost, so we must keep retrying and
-                    # wait for the winner to populate the value.
+                    # wait for the.flags.winner to populate the value.
                     if i < lease_policy.miss_retries:
                         continue
                     else:
                         # We run out of retries, behave as a miss
-                        return None, result.cas_token
+                        return None, result.flags.cas_token
                 else:
                     # There is data, either the is no lease or
                     # we lost and should use the stale value.
-                    return result.value, result.cas_token
+                    return result.value, result.flags.cas_token
             else:
                 # With MISS_LEASE_TTL we should always get a value
                 # because on miss a lease empty value is generated
@@ -378,7 +378,7 @@ def get_cas(
         if result is None:
             return None, None
         else:
-            return result.value, result.cas_token
+            return result.value, result.flags.cas_token
 
     def _get(
         self: HighLevelCommandMixinWithMetaCommands,
@@ -414,7 +414,7 @@ def _process_get_result(
     ) -> Optional[Value]:
         if isinstance(result, Value):
             # It is a hit
-            if result.win:
+            if result.flags.win:
                 # Win flag present, meaning we got the lease to
                 # recache the item. We need to mimic a miss, so
                 # we set the value to None.