[Tracking] 0.5 Release

[MichaelBroughton]

Cirq 0.10.0 is coming sometime in March. Based on past releases TF 2.5.0 should also be coming around that time. We should cut our latest 0.5.0 that depends on both of these. Here are the features we want for the release (assigned people will deliver on them):

1. Implement tfq.noise module. Support for Quantum Channels #250 . Implement sample, sampled_expectation and expectation. Owner: @MichaelBroughton
2. Natural gradient op (not necessarily differentiator implementation) xref [Discussion] Adding quantum math features #380 Owner: @jaeyoo
3. Inner product op gradients + Fidelity op xref [Discussion] Adding quantum math features #380 Owner: @jaeyoo
4. Differentiators get circuits support. Rest of the PRs in this chain here Gradient circuits 1/n: interface #409 , Gradient circuits 2/n: LinearCombination #429 Owner: @zaqqwerty
5. Noise tutorial finished Noise tutorial #454 (happens after, not needed for release) @MichaelBroughton , @zaqqwerty
6. Upgrade to Cirq 0.10, TF 2.5.0 Owner: @MichaelBroughton
7. Release python 3.9 builds if TF will support it. Owner: @MichaelBroughton

Let's get ready for TFQ 0.5.0!

[zaqqwerty]

Will also need to upgrade tutorials, we should pin versions of TF and TFQ, see #555

[MichaelBroughton]

We are quickly approaching our 0.5.0 release. All we are really waiting on now is TF 2.5.0 and py39 builds in order to release. I'm going to quickly document some of the PRs here:

1. For noise related functionality we had:
   Fixes typo breaking ODR on import quantumlib/qsim#308, [Noise] Adds errors for noiseless ops. #500, [Noise] Adds noisy versions of util circuit functions. #501, Added noisy analytical expectation. #508, Adds tfq.noise.samples op. #518, Removed DM from gradient_test.py #520, Add tfq.noise.sampled_expectation. #531, Added docstring for noisy expectation. #536, Added noisy backend to sample layer. #544, Update ***PQC layers to use new backend conventions. #547
2. We decided to cut natural gradient operation from this release and will push it back to a later date.
3. For fidelity and inner product we had:
   Fidelity op #554, Update inner_product_op.py #474 and Add adjoint differentiation of tfq.math.inner_product() #477
4. For the differentiator seperation we had:
   Gradient circuits 1/n: interface #409, Gradient circuits 2/n: LinearCombination #429, Gradient circuits 2/n: LinearCombination #522, Gradient circuits 3/n: implementation for ParameterShift #532, Gradient circuits 4/n: update LinearCombination #533 and Gradient circuits 5/n: refactor base Differentiator #537
5. Nois tutorial is still underway and pending review here Noise tutorial #454
6. Cirq was bumped here: Bump cirq to 0.10.0. #498
7. We are still waiting on TF 2.5.0 and py 3.9 coming with them. ( Will post more on this issue once the outstanding points are finished.)