Cleanup redundant code for computing t complexity

[anurudhp]

Cleaning up some redundant code to compute t-complexity.

• remove _get_all_rotation_types (in favor of bloq_is_rotation).
• reduce uses of t_counts_from_sigma (in favor of QECGatesCost).

TODOs in follow up:

• Currently, only t_counts_from_sigma uses eps correctly, but it's only used in a handful of places. This should be replaced with QECGatesCost after adding support for eps (Propagate rotation epsilon in gate cost #1250).
• We could add a simple GateCounts.total_t_and_rotation_count helper to support existing use cases which only take number of rotations (e.g. Beverland-et-al model).

Relevant discussions:

• t_complexity rotations  #662 (comment)
• Unify t-complexity protocol #678 (comment)

[mpharrigan]

Generally very reasonable changes; unsure about that one test deletion