handling runtimes

[HelenaLC]

• tradeoff b/w economical vs. meaningful content (e.g., using "real life" datasets)
• some methods (e.g., deconvolution, workflows) take looong, even with subsampling
• we could stash precomputed results on OSF, but this defeats the purpose
  of dynamic content (risk of being outta synch and non-reproducible)
• it'd be nice if we should track the timing of all things
  to help identify problematic code chunks, chapters etc.
• how did OSCA handle this? do we need to "just deal with it"?
• can we parallelize the build? at least seq- and img- are fully independent,
  so are workflows... the BBS won't put up with it, but it'd help during development