Prototype: change credit scaling, using nominal vs physical credit

[dcoutts]

Previously we've had a somewhat complex method to scale credits from
those supplied in a level (1 per update) to the credits used in merging.

The existing scheme has a number of downsides:

• no clear distinction between scaled and unscaled credits, what does
  each one measure?
• complex, needs scaling dependent of the merge-policy
• always uses worst case supply of credits so often finishes early
• rounding errors compound problem of credit over-supply
• no satisfactory way to check we do not over-supply credits
• contributing credits to a merge from multiple handles will cause the
  merge to finish earlier than any of the handles needs it to finish,
  thus doing work more eagerly than necessary.

The new scheme involves:

• distinguishing physical vs nominal credits, with a clear definition
  for each measure
• converting between nominal and physical without merge policy
• physical debt is no longer worst case but matches actual cost
• integer rounding errors do not compound
• we can assert that we reach the nominal and physical debt totals at
  the same moment (when the merge is done) and thus we can assert no
  over-supply of physical credits.
• we can avoid supplying credits too quickly to a merge that is shared
  between multiple handles, each one only pushes it as far as it needs.

[mheinzel]

Nice.

[dcoutts]

@mheinzel PR review applied in two extra patches. If you think that's good now then I'll squash and merge.