long running sessions

[Qian Yun]

I need some benchmark on long running sessions to determine
how much speedup can be gained from my kernel cache recycle patch.

During the process, I find the system slows down after many inputs,
more specifically:

After 50k integrals, it is noticeably slower, for anything you
input, it feels like a 200ms lag.

Profiler points to "writeHistModesAndValues".

Because integration implicitly creates algebraic number
(rootOf/nthRoot), and those %%xyz variables are implicitly
added to interpreter. After 50k integrals, there's 10k of
those variables.

I have disabled history and undo, so this
"writeHistModesAndValues" feels unnecessary.
)set history off
)lisp (setq |$undoFlag| nil)

Also its implementation feels questionable.

The history related mechanism is a bit messy.

- Qian