i.e. when applying a case->, the result is the intersection of
each applicable range.

Here are some quickly gathered performance results from this change:

Compile time ratio (old time / new time):
schml-specify-rep: 0.99 (i.e. slower, 8.63s (σ 0.09) to 8.72s (σ 0.05))
schml-interp-casts-help: 1.01 (i.e. faster, 20.05s (σ 0.24) to 19.77s (σ 0.01))
parser: 0.95 (i.e. slower, 2.22s (σ 0.05) to 2.33s (σ 0.09))
old-metrics: 0.96 (i.e. slower, 2.68s (σ 0.04) to 2.78s (σ 0.09))
new-metrics: 0.98 (i.e. slower, 3.61s (σ 0.02) to 3.7s (σ 0.08))
math-flonum: 0.98 (i.e. slower, 3.76s (σ 0.02) to 3.83s (σ 0.08))
fsm: 0.94 (i.e. slower, 4.91s (σ 0.09) to 5.22s (σ 0.19))
forth: 0.97 (i.e. slower, 4.87s (σ 0.18) to 5.01s (σ 0.01))
dungeon: 0.95 (i.e. slower, 8.64s (σ 0.11) to 9.11s (σ 0.18))
bernoulli: 0.95 (i.e. slower, 3.95s (σ 0.04) to 4.16s (σ 0.09))
acquire: 0.96 (i.e. slower, 15.53s (σ 0.23) to 16.2s (σ 0.08))

And for the math library, compile time increased from 2m28.281s
to 2m40.492s.

Note that some primtives (i.e. like those which operate on the
numeric tower) currently have quite verbose function types which
can be greatly simplified after this change is adopted.