order of arguments to build_xxx_kernel is arguably confusing (stylistic issue)

[murphyk]

The API for build_xxx_kernel has the form

build_kernel(varargs,  loglikelihood, logprior, data, batch_size, kwargs)

where varargs is a variable number of arguments, depending on the kernel.
For example

 build_sgld_kernel(dt, loglikelihood, logprior, data, batch_size)
 build_sgldCV_kernel(dt, loglikelihood, logprior, data, batch_size, centering_value):
build_sghmc_kernel(dt, L, loglikelihood, logprior, data, batch_size, alpha=0.01, compiled_leapfrog=True)
...

Arguably it might be nicer to put varargs after batch_size, so the first 4 args are always the same, so that if we add methods like stan_warmup, that automatically estimate dt, we don't have to specify it as the first arg. eg.

 build_sgld_kernel(loglikelihood, logprior, data, batch_size, dt)
 build_sgldCV_kernel(loglikelihood, logprior, data, batch_size, dt, centering_value):
build_sghmc_kernel(loglikelihood, logprior, data, batch_size, dt,  L, alpha=0.01, compiled_leapfrog=True)
build_nuts_kernel(loglikelihood, logprior, data, batch_size, num_warmup)
...

[jeremiecoullon]

This is a good idea! I'll fix this soon.