Hello Guilers!

Following our discussion on the unpleasant static initialization code
(in the `bdw-gc-static-alloc' branch) for subrs because of the dichotomy
between "simple subrs" (with few arguments) and "generic subrs" (with an
arbitrary number of arguments), I investigated all this.

Currently, "generic subrs" (or "gsubrs") are implemented using "compiled
closures" (or "cclos"), as can be seen in `create_gsubr ()'.  Compiled
closures are essentially a wrapper (a cell) around a zero-argument subr
that conveys information about the real number of required, optional,
and rest arguments.

Prior to the switch to double-cells as the storage unit for subrs, the
24 MSBs of the type tag of a subr were used to store the "subr table
entry number" of that subr.  Now that we no longer use a table, those 24
bits are unused.

The attached patch creates a new type tag, `scm_tc7_gsubr', whereby the
24 MSBs are used to store gsubr arity information as returned by
`SCM_GSUBR_MAKTYPE ()'.  This makes cclos useless, which simplifies the
code and reduces the overhead when creating and invoking such
procedures.

In theory all subrs could be encoded as gsubrs, but the interpreter
would need to be able to handle them as efficiently as the specialize
`scm_tc7_subr*', and fiddling with the interpreter may not be so useful
these days.

Objections to committing this?

Thanks,
Ludo'.