Speed up the override registration phase

[robdockins]

by prefiltering possible overrides first.

This significantly reduces the amount of startup time that is consumed
by installing overrides, especially C++ overrides that perform name demangling.

Each override that wants to be registered now provides a "matcher" which examines
the name of a symbol for an approximate test as to weather this override applies.
These can either be exact matches, prefix matches, or substring matches. Any
override that passes the prefilter test is considered in order, as before, to
determine more precicely if it applies.

On one modestly-sized example, this reduced overall verification time from
105 seconds to 32 seconds. The algorithm used here for pre-filtering is
quite naive, so additional improvements are perhaps possible. However,
at the momemnt, startup time is dominated by parsing the bitcode file itself.

Partially addresses #155

[langston-barrett]

This seems like a straightforward and pragmatic approach.

[langston-barrett]

Huh, the CI failure looks unrelated:

Configuring boomerang-1.3.1...
Preprocessing library for boomerang-1.3.1..
Building library for boomerang-1.3.1..
[1 of 9] Compiling Text.Boomerang.HStack ( Text/Boomerang/HStack.hs, dist/build/Text/Boomerang/HStack.o )
[2 of 9] Compiling Text.Boomerang.Pos ( Text/Boomerang/Pos.hs, dist/build/Text/Boomerang/Pos.o )
[3 of 9] Compiling Text.Boomerang.Prim ( Text/Boomerang/Prim.hs, dist/build/Text/Boomerang/Prim.o )
Text/Boomerang/Prim.hs:57:10: error:
    • No instance for (Applicative (Parser e tok))
        arising from the superclasses of an instance declaration
    • In the instance declaration for ‘Monad (Parser e tok)’
   |
57 | instance Monad (Parser e tok) where
   |          ^^^^^^^^^^^^^^^^^^^^

[robdockins]

That CI failure is really strange... it's trying to build a really old version of boomerang.