simulate disk fulls in copyBytes [LUCENE-2726]

In #3711, i disabled copyBytes optimization (so it just calls writeBytes), but i mentioned there,
that I think it would be good to try to beef up tests if we ever wanted to have an optimization like this.

the problem was that when there was an index corruption bug, it was very difficult to detect with our tests.
So I think for safety, we should do this, even though its redundant with our current impl, since we look
for this in writeBytes.

---
Migrated from [LUCENE-2726](https://issues.apache.org/jira/browse/LUCENE-2726) by Robert Muir (@rmuir), updated May 09 2016
Attachments: [LUCENE-2726.patch](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-2726/LUCENE-2726.patch)

Start by reading the copyBytes and writeBytes implementations and the existing tests around them. Add coverage that simulates disk-full conditions during copying, then run the relevant Lucene tests and verify that corruption is detected reliably.

Written by the indexing model from the issue text.