Add support for asynchronous allgatherTensor.

[gchanan]

Because allgatherTensor is a multi-stage process (allgather the sizes, allgather the data), we
can't easily perform it asynchronously. Instead, we complicated the API a bit to make each step
explicit. That is, instead of allgatherTensor(input, output) the API is now:

local tensordesc = mpi.newTensorDesc(mpi.size())
mpi.allgatherTensorDesc(input, tensordesc)
mpi.resizeTensorFromDesc(output, tensordesc)
local handle = mpi.allgatherTensor(input, output, tensordesc)
mpi.freeTensorDesc(tensordesc)

Note also that resizeTensorFromDesc can result in the storage for the output changing
(in case the provided storage is too small).

This could be improved to handle new/free internally, and to provide a single-function
synchronous version.

[facebook-github-bot]

Thank you for your pull request and welcome to our community. We require contributors to sign our Contributor License Agreement, and we don't seem to have you on file. In order for us to review and merge your code, please sign up at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need the corporate CLA signed.

If you have received this in error or have any questions, please contact us at cla@fb.com. Thanks!

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Facebook open source project. Thanks!