Multiple integrals

[gmarkall]

Support code generation for multiple integrals per form. Changes are:

• Refactor the form backends so that most of the functions are passed the integral as a parameter rather than the entire form, to make it easy to distinguish which integral they're supposed to be working with.
• Add a loop to the compile function of the form backend, that iterates over the form's integrals and calls _compile_integral (which now does most of the work that was done by the compile function).
• Edit the assembler so that it also iterates over integrals in a similar way. This required some re-ordering of function calls in the run_model_ function.

At first it may seem odd that all the parameters of the kernels for each integral in a form take the same parameters, regardless of which coefficients they use. However, this is because the coefficients are stored per-form in the form data. Similar behaviour can be observed in FFC - all of the coefficients must be passed to the tabulate_tensor function even if it doesn't need all the coefficients for that integral.

Because these changes involved modification of the generated code, the changes have been tested first on the buildbot - all tests have passed on the devtests-grm08 and fluidity-dev-grm08 queues.

[kynan]

Great! I also like that building expressions is now handled in a cleaner way (i.e. the case distinction for the local tensor eliminated. Merging...