Release 4.0-beta6

This release decreases memory usage. It also moves MCMC and logging commands into the model script.

• Changes:
  • MAP alignment is no longer printed.
  • the MPD alignment is now shorter and more condensed (it uses the "multiply" criterion).
  • the initial alignment has all sequences left-aligned when estimated (for now).
  • -m Model.hs now looks for a Model.main function that returns IO (Random [(Key,JSON)]).
  • stdout / stderr now go to the screen instead of being redirected to a file.
• Features:
  • Tree-alignment models can be hand-written -- no magic return type.
  • Initial implemention of sequence simulation.
  • new executable mcon-tool can convert JSON logs to TSV.
• Model language:
  • Allow reading / writing trees and alignment from the model script.
  • Run loggers and transition kernels from the model script.
  • Refactor Tree object to allow adding and removing nodes.
• Bug fixes:
  • Memory: reclaim memory from running loggers + transition kernels.
  • Memory: clear tokens immediately - don't wait for garbage collection.
  • Memory: don't retain calculations for branches that might not exist.
  • Alignment: fix alignment with likelihood rescaling.
  • Likelihood: fix SEV likelihood on 2-taxon trees.
  • Typechecking: don't allow e.g (x,y) <- return (x,y,z)
  • Typechecking: handle exporting types without their constructors.
  • machine: don't re-execute side-effects in incremental_evaluate1( ).