This prevents the deadlock reported by John Morales in MGO-57.

This is a refactoring of the first batch + cursor logic that is used
in the Pipe and Repair methods, so it may be used both internally in
other places and also externally when people depend on functionality
not yet implemented in the driver.

Also fixes result.Cursor.NS handling, as mentioned by Jeff Yemin.

Issue reported by Daniel Gottlieb.

Fixed typo.

Fixed small documentation typo on GridFs.Create

Besides user error, that situation may happen when the
synchronization goroutine finds the server unavailable.

Reported by John Morales.

Updates #65.

Also improve the documentation for the index key handling.

Updates #65.

MongoDB only supports a single index with a given index key, so
the custom name is a distraction. It will create surprising behavior
where a given index is not created because a different index with
the same key already exists, and it breaks the use of the current
API, such as with DropIndex which is based on the index key.

Updates #65.

Closes #65.

Fixes #66.

Fixes #67.

Reported by Louisa Berger.

The new function accepts a socket, and will be used by an upcoming
version of writeQuery.

This is temporary, for testing the feature in the v2-unstable branch.
The feature will most likely be enabled on 3.0+ only.

As seen on: http://docs.mongodb.org/manual/tutorial/build-a-geohaystack-index/
the haystack index uses a property called bucketSize, which isn't
currently available through this driver. This commit adds the property

Addresses go-mgo#85 and go-mgo#97.

Adds support for Index.BucketSize.

Fix a case of incorrect stats collection.

Not checking for iterator errors was hiding problems with pipeline
result size limits being exceeded when PurgeMissing was being used to
recover a production database.

First iteration of PurgeMissing improvements by Menno.

Fixes #82.

This handles an edge case in which the "rescanned document
misses transaction in queue" error might correctly show up,
and turns the panic into an error otherwise.

This was never present in a stable release.

Indexes each element of the array rather than allocating a slice to
copy the array into as in go-mgo#74.

Fixes #73.

Under MongoDB 2.4 and earlier, aggregation pipeline results may be no
bigger than 16MB, even when an iterator is used. It was possible for
PurgeMissing to hit this limit when dealing with large txn-queue
arrays, causing it to fail.

This change refactors PurgeMissing so that it no longer uses an
aggregation pipeline, working around the limitation. A regression test
is included which triggered the failure mode for the previous version
of PurgeMissing.

The aggregation pipeline result size limitation was removed in MongoDB
2.6. See: http://docs.mongodb.org/manual/core/aggregation-pipeline-limits/

Fixes go-mgo#113.

The panic in those situations was a slice out of bounds error,
which is safe but not ideal.

Fixes #116.

Fixes #117.

Fixes #120.

Two connection string options (connect=replicaSet and
replicaSet=<setName>) were supported in the code but not documented
alongside the other supported connection string options.

The pinger was being run synchronously to obtain an initial
ping value on mongoServer creation, which meant a large delay
when the server was not available. Now the server is initialized
with a large ping value to penalize it until the background
pinger has a chance to update the value, and the server is
returned immediately.

Reported by John Morales on https://jira.mongodb.org/browse/MGO-54

Fixes #142.

Includes internal improvements to the bulk running.

Makes Go 1.5 happy, and fixes #153.

Make the mode type public.

Makes Go 1.5 happy, and fixes #153.

Update documentation to io.EOF instead of os.EOF

Doing that is generally a mistake, since one may decode such data
straight into a []byte value. But fixing for correctness anyway.

Fixes #157.

Still need to implement full IPv6 resolution properly.

Still need to implement full IPv6 resolution properly.