Add CliffordSimulator to cirq.Sample

[smitsanghavi]

1. cirq.sample calls CliffordSimulator when appropriate.
2. All simulators called by cirq.sample now support 0 repetitions even when measurements are not terminal
3. ClifforSimulator exposes supported gates and supports Param resolution

[smitsanghavi]

#2423

[vtomole]

LGTM

[Strilanc]

This is a good start, but we're going to need some pretty wide ranging changes to enable making this change. We need to define three new protocols:

1. has_stabilizer_effect(self) -> bool

   Operations and other objects can implement _has_stabilizer_effect_ to indicate they are compatible with stabilizer simulation. We should add this to all existing stabilizer effects in Cirq, and also augment the standardized tests for gates to include a test like "if it has a stabilizer effect it better have the _has_stabilizer_effect_ method for efficiency" similar to the test we already have for has_unitary. If an object does not implement _has_stabilizer_effect_ the global has_stabilizer_effect method will fall back to calling or _decompose_into_stabilizer_operations_ or else falling back to calling _decompose_ and checking if all the resulting operations all have stabilizer effects.

2. decompose_into_stabilizer_operations(self) -> List['cirq.Operation']

   Operations and other objects can implement _decompose_into_stabilizer_operations_ to specify a decomposition into operations with stabilizer effects.

   If an object does not implement _decompose_into_stabilizer_operations_, the global decompose_into_stabilizer_operations method will fall back to calling _decompose_.

3. apply_stabilizer_effect_to_ch_form(self, args: 'cirq.ApplyStabilizerEffectToChFormArgs')

   "Atomic" operations that can't be decomposed into other stabilizer effects can specify this method for the Clifford simulator to call. The ApplyStabilizerEffectToChFormArgs would contain both the ch_form attribute and also a measurement record attribute.

We should include a unit test that defines a test gate that has a stabilizer effect via decomposition and via apply, and confirm that it works despite the simulator not knowing about it directly.

[Strilanc]

This check will fail on gates that decompose into Clifford operations, such as cirq.ISWAP.

I think what we need to do here is to define an is clifford protocol as well as formalize the decompose_into_clifford protocol. This line would then reduce to has_clifford(circuit).

[smitsanghavi]

Added a TODO for now

[smitsanghavi]

I agree with the need to implement all the changes you suggested to make this fully work as intended in #2423 . This is just the initial version which aims to get this to start working for the simplest eligible circuits. Added a TODO to document that this approach is not permanent.

[Strilanc]

LGTM as an interim thing