For some time I've been wishing for asynchronous Babel session
evaluation. So I've created an experimental branch implementing this. I
have an initial version working for R, so I thought it'd be a good time
to seek feedback and gauge interest.

To test the attached patch, add ":async yes" to an R session block
with a long computation (or "System.sleep") in it. Upon evaluation,
your Emacs won't freeze to wait for the result -- instead, a
placeholder will be inserted, and replaced with the true result when
it's ready.

I'll note how this is different from some related projects. ob-async
implements asynchronous evaluation for Babel, but it doesn't work with
sessions. emacs-jupyter, ein, and ob-ipython all implement asynchronous
session evaluation, but only for Jupyter kernels. Jupyter is great for
some cases, but sometimes I'd prefer not to use it. For example, the
native R console has great Emacs support via ESS, whereas the Jupyter R
console doesn't work with ESS and is not widely used in the R community.

Note that if you use ob-async, make sure to add "R" to
`ob-async-no-async-languages-alist' before testing this.

The new functionality is mainly implemented in
`org-babel-comint-async-filter', which I've defined in ob-comint.el,
and added as a hook to `comint-output-filter-functions'.  Whenever new
output is added to the comint buffer, the filter scans for an
indicator token (this is inspired by
`org-babel-comint-with-output'). Upon encountering the token, the
filter uses a regular expression to extract a UUID or temp-file
associated with the result, then searches for the appropriate location
to add the result to.

I've tried to make behavior as similar as possible to existing
ob-comint behavior, so that some of the existing code for interacting
with ob-comint can be refactored and reused. Still, it will be a large
task to add this feature for all languages. So far, I've only done R,
but my thought is to implement a few more languages before nailing
down the functionality. But, I hope something like this could be
merged in supporting just a subset of languages initially, then
gradually increasing the number of supported languages over time.