[df] Lift requirement to provide explicit return type in Vary expression

[vepadulano]

While working on the presentation about systematic variations for the ATLAS columnar hackathon https://indico.cern.ch/event/1661946/, I had an idea on how to simplify the user experience of using Vary with JIT-ted expressions.

The expression gets a trailing return type injected of the equivalent necessary RVec type, only in case the user is not already providing it. As long as the expression returns a sequence of characters that is a valid way to instantiate an RVec constructor, the expression will compile.

The condition that triggers the injection of the trailing return type is if the user expression is of type '{...}', which could be a potential candidate constructor for an RVec.

[github-actions]

Test Results

    22 files      22 suites   3d 10h 48m 33s ⏱️
 3 850 tests  3 848 ✅  1 💤 1 ❌
76 910 runs  76 891 ✅ 18 💤 1 ❌

For more details on these failures, see this check.

Results for commit 2d2f03a.

♻️ This comment has been updated with latest results.

[hageboeck]

This is really great to have!

Let's also add a bit of docs, because otherwise it might go unnoticed. Something to the effect of:

/**
df.Vary(
  std::vector<std::string>{"x", "y", "z"},
  "{
    {-1*entry, 2, 3},              // x variations
    {41, 43*entry, 44},            // y variations
    {500-entry, 600, 700 + entry}  // z variations
  }",
  {"down", "up", "other"},
  "xyz")
*/