Updated accelerators with variable timestep, add localization case to examples

[odunbar]

Purpose

• removal of explicit timestep dependency of NesterovAccelerator() appears to be a more robust approach with a do-no-harm adjustment

Content

• removal of Delta t and Delta t_prev dependency
• add fig_save directories and fig_save naming based on timestep for accelerated examples.
• add some info statements, more encapsulated "cases", (i) constant stepper, (ii) dmc stepper (iii) dmc with localization and small ensemble (size = 5)
• Add GC into Darcy example over random trials.
• Added simple ConstantNesterovAccelerator which sets the momentum parameter = constant
• Renamed ConstantStepNesterovAccelerator to FirstOrderNesterovAccelerator
• Fixed an off-by-one error in FirstOrder

Example figures from Accelerated/

• KEY INSIGHT: Acceleration still proves to have increased benefit for small ensembles while using localization (common in practical scientific needs)

ExpSinAccelerated for case = const , case = dmc and case = dmc-loc-small-ens:

LorenzAccelerated for case = const , case = dmc and case = dmc-loc-small-ens:

DarcyAccelerated for case = const , case = dmc and case = dmc-loc-small-ens:

---

• I have read and checked the items on the review checklist.

[eviatarbach]

LGTM!