Remove Scir.cbrain_config stuff; add support for NCPU per tool config.

[prioux]

This structure was meant to hold info about
the default queue and extra qsub args, but instead
let's delegate the responsibility for those values
to the code that launches tasks using scir (in the case
of cbrain, ClusterTask's submit_cluster_job().

That way, we can substitute the ppn=N values from
the tool_config's ncpu attribute, if we agree for
a convention for it (e.g. if the bourreau's
cms_extra_qsub_args attribute contains something like

"-l node=1:ppn={NCPU}"

then submit_cluster_job() would make the substitution
before passing the string to the job object creator.

[natacha-beck]

@prioux I see no reference to Scir.cbrain_config with ncpu.
We use Scir.cbrain_config for the following:

• default_queue
• amazon_ec2_access_key_id
• amazon_ec2_secret_access_key
• amazon_ec2_region
• extra_qsub_args

Should all of this move to tool_config ?

[prioux]

No let me handle all that, there's a fundamental redesign of all this that need to be undertaken.