Improve performance of sensi computations with applied RAs

[pet-mit]

Describe the current behavior

When computing sensitivity values with applied RAs, a network clone is created for every state with applied RAs.

Describe the expected behavior

Improve performance by avoiding the creation of new clone, and instead re-using existing clones (but we will have to un-apply RAs in-between states). (see TODO in SystematicSensitivityAdapter around line 111)

Describe the motivation

Performance boost

Extra Information

No response