Arrow one

[mtzguido]

arrow_one_ln was needlessly closing over the binders again, even though they were already closed. This is a no-op, but it makes some programs quadratic instead of linear, for instance if calling arrow_one_ln repeatedly from a tactic, which happens for collect_arr_bs, and others.

For an arrow with 500 binders and the mk_projectors tactics, this reduced the startup time from ~3s to 180ms. There's other bottlenecks probably.

[mtzguido]

Running a full check here: https://github.com/mtzguido/FStar/actions/runs/11039760857