Add a do-block form of the CachingPool

[oxinabox]

This feels right.

Generally I think there is some region of code in which the cached functions need to be keep.
So a do-block form that runs the function,
then when done clear!s the pool seems right.

I guess maybe not required as the finalizer will clean up once the pool get garbage collected.
But given that the reason you are using a CachingPool in the first place is because you are worried that you have huge closures being passed around,
clearing sooner is better than latter.

This needs tests I guess, but I figured I'ld wait and see if the idea was sound

[oxinabox]

A.V has 64 bit failure with

Please submit a bug report with steps to reproduce this fault, and any error messages that follow (in their entirety). Thanks.
Exception: EXCEPTION_ACCESS_VIOLATION at 0x680b3d78 -- gc_mark_loop at /cygdrive/c/projects/julia/src\gc.c:1858
in expression starting at C:\projects\julia\usr\share\julia\stdlib\v0.7\Pkg\src\precompile.jl:75
gc_get_stack_addr at /cygdrive/c/projects/julia/src\gc.c:1351 [inlined]
gc_read_stack at /cygdrive/c/projects/julia/src\gc.c:1358 [inlined]

during precompilation.

CircleCI failed during make for reasons that I can't see displayed

[martinholters]

The AV failure is probably because it was run prior to #27702, the "Killed" on Circle is usually OOM, IIUC.