perf: stop simplifying the slow-flow equations twice before averaging

[oameye]

Depends on QuantumEngineeredSystems/QuestBase.jl#82 (block-diagonal solve and the declare_variable change).

get_krylov_equations ran Symbolics.expand.(Symbolics.simplify.(...)) twice over the slow-flow equations, on the grounds that once was not enough to fully simplify them. What the second pass actually bought was the cancellation of the denominator rearrange! leaves behind, and Symbolics.simplify is a poor way to get it: its cost explodes with the number of harmonics.

average works term by term, so a flat sum of products is all it needs. A single expand gives that. The denominator is cancelled once at the end with simplify_fractions, after averaging has removed the fast time and the expressions are small.

take_trig_integral multiplied its numerator by ω and divided the denominator by the same ω, then called simplify to undo it. Termwise integration leaves the denominator alone, so it goes back as it came.

Together with the QuestBase changes, get_krylov_equations over the test systems goes from 1412 MiB to 925 MiB, and get_harmonic_equations from 351 MiB to 164 MiB.

The precompile workload is re-enabled (it was commented out because declare_variable's @eval made it impossible) with the two get_krylov_equations calls added.