As there is one RLoopManager per computation graph, when aliases
were managed by RLoopManager they were computation-graph-wide.
It is desirable to make Alias definitions behave coherently with
Defines instead, i.e. have Aliases be only accessible in the branch
of the computation graph in which they were defined, and only in
nodes that are downstream of the one where the alias is added.

This resolves root-project#7381, "[DF] Let Aliases be defined per computation graph
branch, not globally".

In particular, in this commit:
- move alias management from RLoopManager to RBookedColumns
- remove alias-managing logic from RLoopManager
- refactor several functions so they only take RBookedColumns as input
  rather than a list of defined names plus a map of aliases
  (CheckForDefinition, CheckForRedefinition, FindUnknownColumns,
  GetValidatedColumnNames, BookFilterJit, FindUsedColumns,
  ParseRDFExpression)
- use RBookedColumns::ResolveAlias instead of ResolveAlias helper function
- adapt dataframe_utils test to the new signature of FindUnknownColumns

Note that this is a backward-incompatible change for user code that
relied on this global definition of Aliases. We expect this to be an
exceedingly rare usecase, and if such code exists it will produce a
clear error message.

RBookedColumns has a better API now, and client code does not need
to "manually" add names to its internal list.

It used to be impossible to Define (or Redefine, or Alias) a column
that had already been Aliased from anywhere in the computation graph.

Now that Aliases are specific to a branch of the computation graph,
we can add a certain Alias in one branch and still Define a column
with the same name in another.