make the pre-[hg update] checks faster (part 2)

This contains the more intrusive/less impactful optimizations that I left for later. This is much less important than !388 (merged), so please prioritize that if you have to choose when reviewing.

Benchmarks go something like this (this time on a real FS instead of a ramdisk, so a bit worse and noisier than the previous bench):

Benchmark: hg update tip --config devel.debug.abort-update=1
s0. Time (mean ± σ):     11.832 s ±  0.137 s
s1. Time (mean ± σ):      8.338 s ±  0.086 s
s2. Time (mean ± σ):      8.295 s ±  0.117 s 
s3. Time (mean ± σ):      8.210 s ±  0.031 s
s4. Time (mean ± σ):      5.828 s ±  0.077 s
s5. Time (mean ± σ):      5.308 s ±  0.043 s
s6. Time (mean ± σ):      5.207 s ±  0.033 s
s7. Time (mean ± σ):      4.921 s ±  0.049 s