Narrow the tuning space of sq auto-tune

[yiliu30]

Type of Change

API changed or not: None

Description

After trying all alpha values, the sq tuning process will stop early, returning the current best qmodel, even if the current best accuracy does not meet the accuracy criterion.
For example:

alpha = [0.1, 0.5]
acc_baseline = 1
acc_alpha1 = 0.8  # alpha is 0.1
acc_alpha2 = 0.9  # alpha is 0.5
# continue tuning process with op-type wise tuning/fallback
# Change to ->
# end the tuning process, return the qmodel with alpha is 0.5

How has this PR been tested?

Pre-CI

Dependency Change?

None