Fix `NEB` so that it works with `method != 'aseneb'` and `remove_rotation_and_translation == True`

Related to #1642

This MR solves the issue I described in #1642 (comment 2423398615).

Specifically, what happened are

• When remove_rotation_and_translation == True, the positions of the last end point in NEB is modified.
• This discards atoms.calc.results in the end-point configurations.
• Particularly for SinglePointCalculator, there is no way to recalculate them.
• This causes an issue here with `method != 'aseneb', where the end-point energies are accessed.

The present MR solves the issue above by storing the end-point energies at the initialization of NEB, before the modification of the positions due to remove_rotation_and_translation == True.

This is tested by extending test_neb_tr for the methods other than aseneb.

Checklist: x (done) or ~ (irrelevant)

• I am familiar with ASE's contribution guidelines.
• Doc strings in code changed in this MR are up to date.
• Unit tests have been added for new or changed code.
• Changes have been added in changelog.d using scriv.
• Issue is resolved via "closes #XXXX" if applicable.