improve augur_align performance

[dpark01]

Possibly replace augur align invocation with our own mafft-based wrappers that take advantage of recent optimizations. Keep existing WDL task for augur align (in case we want it later) but add a new task (or modify an existing one) to interhost.wdl and use that instead in the main nextstrain workflow(s)

[tomkinsc]

One question is whether we should pursue changes to augur upstream to add something like a --consider_conserved param to augur align that toggles a code path with mafft --addfragments and --keeplength.

[dpark01]

Well one question is whether augur align does much other than a specific incantation of mafft. If it doesn't, maybe the best path is just to call mafft ourselves. But if it's doing something special and nextstrainy, pursuing upstream changes may make sense.

[dpark01]

Quick note, #112 tangentially addressed this a bit by throwing some more CPU at it (since mafft appears to make good use of available cores).