Expose benchmark params dataclass stub

This is intended as a base class holding benchmark parameters. Users
should subclass this to their needs. Also allows a stronger-typed
interface for `params` arguments, e.g. in `BenchmarkRunner.run()`.

src/mlbench/__init__.py

@@ -9,4 +9,4 @@
     pass
 
 # TODO: This naming is unfortunate
-from .core import Benchmark, benchmark, parametrize
+from .core import Benchmark, Params, benchmark, parametrize

src/mlbench/core.py

@@ -10,6 +10,20 @@ def NoOp(**kwargs: Any) -> None:
     pass
 
 
+# TODO: Should this be frozen (since the setUp and tearDown hooks are empty returns)?
+@dataclass(init=False)
+class Params:
+    """
+    A dataclass designed to hold benchmark parameters. This class is not functional
+    on its own, and needs to be subclassed according to your benchmarking workloads.
+
+    The main advantage over passing parameters as a dictionary is, of course,
+    static analysis and type safety for your benchmarking code.
+    """
+
+    pass
+
+
 @dataclass(frozen=True)
 class Benchmark:
     """