Support variable % of clients and relays using a differenet tor build

We need:

• Test networks containing 2 different builds of tor (with and without the new congestion control)
• For these experiments we can have all relays using the same build (the new one), and a variable % of clients using the new build.
• For meaningful results we're going to need multiple trials, which tornettools can use to draw error bars.

We'll add new pipeline variables:

• LEGACY_TOR_REPO - git repo from which to get the "legacy" version
• LEGACY_TOR_BRANCH - git branch from which to get the "legacy" version
• PERCENT_LEGACY, taking a list like [0,50,100]
• NUM_TRIALS EDIT: Multiple trials has been implemented for a while (though it's hard coded @ 3)

e.g. with PERCENT_LEGACY=[0,50,100] and NUM_TRIALS=3 we might end up with a tree of simulations like:

• 0-percent-legacy
  • trial1
  • trial2
  • trial3
• 50-percent-legacy
  • trial1
  • trial2
  • trial3
• 100-percent-legacy
  • trial1
  • trial2
  • trial3

And graphs would have 3 lines - one for each of 0, 50, and 100 percent legacy.