There's some problems with the available QC tools, at least for the
purposes of this project. Namely:

 * the Rust testing framework limits the amount of concurrency
   available, starving some tests of runtime
 * there is no distribution available
 * test-case generation can be slow

Considering we're randomly searching through a state space the name of
the game is speed. To that end, I have in mind a notion of
QuickChecking which operates first in fuzzer mode--quickly throwing
random nonsense into your program to find crashes, track branch
conditions in the program--and, once a crash has been found, switches
into QC mode to shrink the found case. This commit does not have that
mode operation notion, only shuffles code around to run a property
test under AFL. There is no shrinking.

Signed-off-by: Brian L. Troutwine <brian@troutwine.us>

This commit adjust the fuzzer from AFL to honggfuzz. Honggfuzz has the
advantage of being multi-threaded by default but, honestly, the
initial big draw was easy hook-in to lldb. The AFL fuzzer from the
previous commit 'detected' HashMap panicing when we requested too much
capacity but I found that an easy call to "cargo hfuzz run-debug
hash_map" made the failure much more clear.

The main change here is around the Reserve operation, limited now to a
smallish reservation bump. This avoids the case where HashMap will
panic on call to its `reserve`, being unable to signal that its
allocation failed. The model is also careful about requested too much
capacity, as noted. If HashMap published its overhead per pair we
could calculate when the allocations would fail but it, reasonably,
does not publish such information.

Signed-off-by: Brian L. Troutwine <brian@troutwine.us>

I neglected to remove quickcheck from `lib.rs`, though it is no longer
a project dependency. This caused test build of the project to fail.

Signed-off-by: Brian L. Troutwine <brian@troutwine.us>

This commit introduces FiniteBuffer into the hash map
fuzz. FiniteBuffer, unlike RingBuffer, does not produce an infinite
amount of data but will limit production to the total number of bytes
available in `data`. This change means we have to guard each Arbitrary
and it's possible that if the fuzzer is not correctly configured we'll
only check small total operations. But, it is quite a bit faster and
fuzzers that do minimization -- like AFL -- are able to eat into the
shrink step needed for proper quickchecking.

This is promising, I think.

Signed-off-by: Brian L. Troutwine <brian@troutwine.us>

Clippy has changed its lint names to scoped form. This plays with
CI pipelines using stable release until 'tool_lint' feature lands.
Seems like this is targeted for the 2018 edition so, uh, this
change will surely be reverted in the near term.

Signed-off-by: Brian L. Troutwine <brian@troutwine.us>