Deletes a bunch of code and speeds up lowering time of local laplacian
with 20 pyramid levels by ~2.5%

It was O(n) for n facts. This makes it O(log(n))

This was particularly bad for pipelines with lots of inputs or outputs,
because those pipelines have lots of asserts, which make for lots of
facts to substitute in.

Speeds up lowering of local laplacian with 20 pyramid levels (which has
only one input and one output) by 1.09x

Speeds up lowering of the adams 2019 cost model training pipeline (lots
of weight inputs and lots outputs due to derivatives) by 1.5x

Speeds up resnet50 (tons of weight inputs) lowering by 7.3x!

…o abadams/faster_substitute_facts

…o abadams/faster_substitute_facts

…o abadams/faster_substitute_facts

Interval::is_single_point() used to only compare expressions by shallow
equality to see if they are the same Expr object.

However, bounds_of_expr_in_scope is really improved if it uses deep
equality instead, so it has a prepass that goes over the provided scope,
calls equal(min, max) on everything, and fixes up anything where deep
equality is true but shallow equality.

This prepass costs O(n) for n things in scope, regardless of how complex
the expression being analyzed is. So if you ask for the bounds of '4'
say in a context where there are lots of things in the scope, it's
absurdly slow. We were doing this! BoxTouched calls
bounds_of_expr_in_scope lots of times on small index Exprs within the
same very large scope.

It's better to just make Interval::is_single_point() check deep
equality. This speeds up local laplacian lowering by 1.1x, and resnet50
lowering by 1.5x.

There were also places where intervals that were a single point were
diverging due to carelessly written code. E.g. the interval [40*8,
40*8], where both of those 40*8s are the same Mul node, was being
simplified like this:

interval.min = simplify(interval.min);
interval.max = simplify(interval.max);

Not only does this do double the simplification work it should, but it
also caused something that was a single point to diverge into not being
a single point, because the repeated constant-folding creates a new
Expr. With the new is_single_point this matters a lot less, but even so,
I centralized simplification of intervals into a single helper that
doesn't do the pointless double-simplification for single points.

Some of these shallowly-unequal but deeply-equal Intervals were being
created in bounds inference itself after the prepass, which may have
been generating suboptimal bounds. This change should fix that in
addition to the compile-time benefits.

Also added a simplify call in SkipStages because I noticed when it
processed specializations it was creating things like (condition) ||
(!condition).

…_single_point

This visitor shows up as the main cost of lowering in very large
pipelines.

This visitor is for tracking which lets are actually used for real
inside the body of a let block (as opposed to the tracking we do when
mutating, which is approximate, because we could construct and Expr that
uses a Var and then discard it in a later mutation).

The old implementation made a map of all variables referenced, and then
checked each let name against that map one by one. If there are a small
number of lets outside a huge Stmt, this is bad, because the data
structure has to hold a number of names proportional to the stmt size
instead of proportional to the number of lets.

This new implementation instead makes a hash set of the let names, and
than traverses the Stmt, removing names from the set as they are
encountered. This is a big speed-up.

We then make the speed-up larger by about the same factor again doing
the following:

1) Only add names to the map that might be used based on the recursive
mutate call. These are very very likely to be used, because we saw them
at least once, and mutations that remove *all* uses of a Var are rare.

2) The visitor should early out when the map becomes empty. The let
variables are often all used immediately, so this is frequent.

Speeds up lowering of local laplacian by 1.44x, 2.6x, and 4.8x
respectively for 20, 50, and 100 pyramid levels.

Speeds up lowering of resnet50 by 1.04x. Speeds up lowering of lens blur
by 1.06x

…sed_in_simplify_let