"I have read the QoS SRND and wanted to ask about the WAN aggregator (ie Head Office/Data Center router). Is it possible to shape per branch?"

That depends, but often the answer is yes.

"The WAN aggregator has a 40Mb connection but a lot of the branch sites are only 2Mb. Is it feasible to configure a policy for every branch so that you shape to the branch connection/provider CIR?"

Another that depends, but sometimes not only feasible but a really good idea if you wish to manage congestion.

"Can this be done as well as shaping the main interface to 40Mb and also marking/classifying different types of traffic such as Citrix is allowed policed to 1Mb for a site and marked down if above?"

Once again, another that depends. Multiple level shaping, I recall, not generally supported (at least directly within "usual" CBWFQ). Subordinate policing might be supported, but wouldn't advise for traffic like Citrix. (Generally, I'm not very keen on using policing ever, alhough sometimes it's all that's available to work with.)