Fix VQC training with warm_start=True

[declanmillar]

Summary

Fixes the extraction of the initial_point in TrainableModel from the final point of the
minimization in the existing OptimizerResult object. Closes #296.

Adds unit tests to ensure VQC correctly handles warm starts.

Details and comments

As pointed out by @woodsp-ibm, "In Terra the optimizers were refactored and the old optimize method, that returned a tuple (subscriptable) was deprecated and a new method minimize was created that returns OptimizerResult object."

This is a very small fix to reflect that change and some related unit tests that would have caught the issue.

[coveralls]

Pull Request Test Coverage Report for Build 1829125173

• 1 of 1 (100.0%) changed or added relevant line in 1 file are covered.
• 3 unchanged lines in 1 file lost coverage.
• Overall coverage decreased (-0.06%) to 83.64%

---

Files with Coverage Reduction	New Missed Lines	%
qiskit_machine_learning/datasets/gaussian.py	3	35.0%

Totals
Change from base Build 1829048709:	-0.06%
Covered Lines:	2817
Relevant Lines:	3368

---

💛 - Coveralls

[woodsp-ibm]

Can you add a reno release note with a fixes section. This is required for issues that impact/affect the public API.

Thanks for adding a unit test that checks this. This is something that we always want if we find our CI has a hole like this case where the tests did not catch it. I see you have modelled it with data after what I see is done elsewhere here. I will note that that data values could be more directly coded into the data e.g directly have an instance of the optimizer etc rather than having this indirect via string values. ddt can also input to multiple parameters and do unpacking via decorator so the test does not need to have that. I am not asking for anything to be changed merely pointing this flexibility out that you may want to look at going forwards if/when using ddt. Eg from our former Aqua code https://github.com/Qiskit/qiskit-aqua/blob/c1564af8792c6664670807614a378147fd04d28f/test/aqua/test_vqe.py#L126-L131

[declanmillar]

This PR should be labelled as Changelog: Bugfix and stable backport potential. I don't appear to have the necessary permissions to apply these myself.