More info shuffling and refactoring

[elibarzilay]

• The code for a dangerLevel of MultiplePackagesEdited was broken:
  it used noNulls over a list of booleans, which means that
  MultiplePackagesEdited corresponded with just having multiple
  packages (not taking into account tests or criticals). This changes
  the code to use just a simple count of packages for now.

  I tried to fix it, but that exposes a problem with dangerLevel in
  general: it is sometimes used as an indicator for a single package
  literally (eg, allow all owners to request a merge if it starts with
  "Scoped") and sometimes as a generic danger situation. This is likely
  my fault since pretty much any use of it confused me (since it's
  defined as a hard-to-remember combination of different things).

  For the same reason, I think that it'd be best to try to remove it and
  use the information directly. Some of the following changes are going
  in this direction. I will revisit the issues around it when this is
  done (eg Improve "multiple packages" label? #206 / Don't take test files into account when looking at whether a package affects multiple packages #155).

• Use an extended editsInfra field instead of dangerLevel === "Infrastructure".

• Working toward the above, move the getDangerLevel computation (and
  the dangerLevel field, and the DangerLevel type) to
  ExtendedPrInfo since there are no more uses of it in pr-info. The
  next step is to split it into fields that express the relevant parts
  instead of a single level value.

• Move authorIsOwner from a PrInfo field to a computed
  ExtendedPrInfo one. It's meaning is also changed: previously, it
  would be on if the author is an owner of any package, and now it's
  on if the author is an owner of all packages (if more than one).
  This currently affects only the corresponding label. Two
  tests (45836, 45946) updated for this.

• Aggregate all review info into a single getReviews (much revised
  from analyzeReviews) that returns a review: ReviewInfo[], holding
  all of the relevant information. Remove all of the other fields, and
  add the ones that are needed into the ExtendedPrInfo.

  There are some minor diffeences --- for example, previously all stale
  reviews were kept, and now it's only one review kind per person so at
  most one stale review. These differences only affect the derived
  information, but no test changes.

• Note that the computation of approvalFlags moved to extendPrInfo
  (in compute-pr-actions), where it can be changed to adapt it to
  address People should not be able to self-merge until maintainers of all affected packages have approved a change. #178.

[orta]

Nice work, looks like net lines removed 👍🏻