Use new quimb TN string format

[95-martin-orion]

Resolves the "tests are failing" part of #5524.

[viathor]

AFAIR, this is the first time this happened with quimb. I'd consider pinning if it keeps happening and affects us more seriously.

[maffoo]

Thanks for looking into this, @95-martin-orion.

This change will cause tests to fail for people who have older versions of quimb installed locally, which seems less than ideal. I don't think we should be relying on string representations provided by third-party libraries to be stable in our tests.

[95-martin-orion]

This change will cause tests to fail for people who have older versions of quimb installed locally, which seems less than ideal.

This isn't the first time we've done something like this - the usual solution is to announce the change and recommend re-installing Cirq requirements.

I don't think we should be relying on string representations provided by third-party libraries to be stable in our tests.

This is the cost of allowing MPS to use quimb-based state for its string representation. Some alternatives:

• Manually construct a string for the quimb state in Cirq
• Don't reveal the state in the string representation
• Don't test the string representation (likely fails coverage checks)

@maffoo, what are your thoughts? I'll hold off on merging until I have your approval.

[maffoo]

My preference in order:

1. Don't test the string representation (or at least, don't test for exact equality, just test that str(...) works and maybe that it contains info that we control and care about)
2. Don't reveal the state in the string representation (seems less useful for users)
3. Manually construct a string for the quimb state in Cirq (seems like work we shouldn't need to take on)

[maffoo]

Approving so we can merge to unblock CI. We can discuss changes to the tests to avoid relying on stability of quimb strings in the future.