type-contract: add common ctc-cache

[bennn]

Make a cache at the top of mb-core and share it across all passes that might generate contracts.

[samth]

Can you evaluate this somehow? For example, how often do we hit this cache? How often do we miss? Is there a performance impact?

[bennn]

Here's a small evaluation. So far, looks like this cache
does not help much.

First, I measured compile times for the tr-performance
programs. Times are within the noise before & after the cache.

Second, I measured zo sizes for tr-performance and the
programs in #917. A few programs do a little better --- task-6
from 7GUI saves 20KB if I'm interpreting the ls -l output
correctly --- but most are the same before & after.

I'm inclined to close this PR for now and take a closer look
later, for my upcoming "transient" pull request. That adds
a new contract-making pass & should give a stronger signal
about whether this is a good idea.

compile time

benchmark	master	share-cache
acquire	15.44 (σ 0.06)	15.53 (σ 0.1)
bernoulli	3.85 (σ 0.07)	3.9 (σ 0.04)
dungeon	7.32 (σ 0.06)	7.36 (σ 0.04)
forth	4.6 (σ 0.07)	4.61 (σ 0.13)
fsm	4.96 (σ 0.04)	4.97 (σ 0.03)
math-flonum	3.79 (σ 0.02)	3.8 (σ 0.02)
new-metrics	4.1 (σ 0.03)	4.11 (σ 0.02)
old-metrics	2.68 (σ 0.03)	2.67 (σ 0.02)
parser	2.33 (σ 0.02)	2.34 (σ 0.01)
schml-interp-casts-help	22.32 (σ 0.4)	22.45 (σ 0.12)
schml-specify-rep	8.53 (σ 0.05)	8.53 (σ 0.07)

zo size

benchmark	master	share-cache
acquire	1.5M	1.5M
bernoulli	248K	248K
dungeon	496K	496K
forth	340K	336K
fsm	364K	364K
math-flonum	172K	172K
new-metrics	168K	168K
old-metrics	68K	68K
parser	56K	56K
schml-interp-casts-help	2.4M	2.4M
schml-specify-rep	740K	740K
-------------------------	----------------	----------------
plot2d.rkt	75774B	75765B
plot3d.rkt	83573B	83564B
task-1.rkt	80886B	80888B
task-2.rkt	83109B	83108B
task-3.rkt	93582B	93584B
task-4.rkt	81577B	81575B
task-5.rkt	113172B	113169B
task-6.rkt	783436B	759673B
task-7.rkt	107778B	107780B
gui-types.rkt	159918B	160127B
framework-types.rkt	278388B	273291B
framework.rkt	228828B	228828B
gui/base.rkt	151591B	151591B
pict.rkt	72925B	72925B
http.rkt	353170B	353161B

[samth]

I think we need to finish #633 and then see how this impacts things.