Fix flaky test due to racy assertion

Fix async refresh from prematurely replacing the value

When a refresh-after-write is triggered the entry should be reloaded in
the background, the present value available for reads, and atomically
replaced when the new value has been loaded. This operation is atomic
with other writes to that entry and is blocking (rather than clobbering)
if another write is attempted.

This should work the same for synchronous and asynchronous caches, but
unfortunately it wasn't. For an asycn cache a new, incomplete future
was immediately put into the cache and available to be consumed by the
next request. Due to layering, the custom reloadAsync was not called and
the operation delegated to load instead. This was of course wrong and
not the intended (or expected) behavior, so it is now fixed.

Thanks to Etienne Houle @ Stingray for notifying me of this problem.

misc cleanup

Prepare for next release

List automatic refreshing in the feature bullet points

Support "no change" in JCache configuration

The expiration policy may return null for update or access queries to
not change the entry's expiration time. This was not supported by the
configuration file, which translated null to mean eternal. This is now
fixed and the default set to "eternal" explicitly.

Add a little tolerance due to async scheduling

When shutting down the executor, other async tasks may be rejected
that are benign to the test. For example the failed future is
removed, a no-op removal notification is published (and ignored),
and a cleanup cycle is triggered (and falls back to the calling
thread if rejected). In these cases the executor can fail without
harming the test.

Fix OSGi manifests; add tests (fixes ben-manes#33)

Release 2.0.0

Allow expired entries to be resurrected (fixes ben-manes#30)

An expired entry may be replaced in the cache prior to it being evicted
by a maintenance cycle. When the maintenace is being performed and reads
a stale timestamp, during eviction is must validate that the entry is
still an eviction candidate. This was reported to have caused excessive
evictions for short time durations, so the race was not benign as
originally thought.