mathlib3

7c0ff89b - perf(analysis/calculus/specific_functions): speed up `exp_neg_inv_glue.f_aux_deriv` (#16812)

perf(analysis/calculus/specific_functions): speed up `exp_neg_inv_glue.f_aux_deriv` (#16812) This caused some [bors failures](https://github.com/leanprover-community/mathlib/actions/runs/3182965873/jobs/5189596416) yesterday. The new proof features a 8x time reduction from >20s to <2.5s on my machine.