Simpler exp val

[aeddins-ibm]

Main changes are in the first commit. Slightly edited version of its commit message:

Previously, first averaged over shots, then averaged over twirls (or other specified axes).

This was complicated by the need to keep track of how many shots were accepted by postselection in order to propagate uncertainty properly later when averaging over twirls. Also by the need to keep track of PEC signs (pauli_signs).

Now, we first flatten avg_axis (typically the twirl axis) into the shots axis, then do all averaging + postselection at once. This avoids needing to work with weights. This should also avoid the situation where postselection rejects all shots on a single twirl, resulting in nan for the average of all twirls (particularly problematic when approaching the theoretically-optimal case of 1 shot per twirl).

[coveralls]

Pull Request Test Coverage Report for Build 23020221462

Details

• 20 of 23 (86.96%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.4%) to 90.244%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit_addon_utils/exp_vals/expectation_values.py	20	23	86.96%

Totals
Change from base Build 22785714859:	-0.4%
Covered Lines:	740
Relevant Lines:	820

---

💛 - Coveralls

[caleb-johnson]

This LGTM, thanks!