Fix #959: count reports the number of Data Points

[albertohernandez1995]

Summary

Applies the meeting decision
on count, and closes #996 with it: the eight cases reported there were checked against the same
decision, as its comment
asked.

What count does now

• A Data Set evaluation applies the operator to each Measure and gives that Measure back counted:
  COUNT(Me_1) AS Me_1. A Data Set holding a single Measure returns it renamed to int_var.
• The Data Set form and the Component form report the same numbers:
  count(DS_1 group by Id_1) and DS_1[aggr n := count(Me_1) group by Id_1] agree.
• A group that counts no value reports 0, not null. The DuckDb transpiler dropped the two
  NULLIF(..., 0) wrappers that turned an empty count into null; the pandas engine fills the
  groups the aggregation leaves out, after the having filter reads them.
• count() with no operand still reports the number of Data Points: it is now given no Measures
  to count, which also fixes it reporting the first Measure's non-null count on the pandas engine,
  and reporting nothing at all when the Data Set is not grouped.

The cases of #996

#	Case	Reported	Now	Settled by
1	count(DS_1)	0	Me_int 9, Me_num 10, Me_str 8	this PR
2	count(DS_1 group by Id_1)	NA	one count per Measure, 0 for an empty group	this PR
3	[aggr n := count(Me_int) group by Id_2]	NA	3, 3, 3, 0	this PR
4	[aggr n := count() group by Id_2]	NA on pandas	3, 3, 3, 1	this PR
5	count(Me_int over (partition by Id_2))	1 on pandas	0	#990
6	count(Me_num over (order by Id_2, Id_1))	1 on pandas	10 on every row	#1002
7	having + viral attribute	KeyError on pandas	3, 3, 3	#990
8	count(Me_str over (partition by Id_2))	TypeError on pandas	3, 2, 2, 1	#990

Two of them answer differently from what that issue expected, on purpose: count over a Data Set
no longer reports the number of Data Points (cases 1 and 2), and an omitted window clause covers
the whole partition (case 6, #1002). Checking the first of those turned up one more bug, fixed
here: DS_1 [ aggr n := count() ] with no group clause raised
Invalid Input Error: Need a DataFrame with at least one column on the pandas engine while
DuckDb answered 10, because an ungrouped count() is left with no column at all.

Checklist

• Code quality checks pass (ruff format, ruff check, mypy)
• Tests pass (pytest, both duckdb and pandas backends)
• Documentation updated (if applicable)

Impact / Risk

• Breaking. count over a Data Set no longer reports the number of Data Points:
  • a Data Set with several Measures returns one counted Measure per Measure instead of a single
    int_var, so [rename int_var to ...] and [keep int_var] stop resolving;
  • a Data Set with a single Measure returns that Measure's non-null values, a smaller number
    wherever the operand holds nulls.
• Grouped counts return 0 where they used to return null.
• Scripts that mean "number of Data Points" migrate to DS [ aggr int_var := count ( ) group ... ]
  or count ( DS#<identifier> group ... ) — an identifier is never null, so both keep the previous
  numbers and the int_var name. Seven test scripts were migrated that way, including the AnaMart
  and AnaVal projects.
• 65 reference files updated to the new shape, and total_records_per_month in GL_466_1 and
  aggr.numDPCouYear in DEMO1 show the effect on real scripts.
• Needs a release note.

Notes

GH_959_1 covers a Data Set of one Measure, GH_959_2 a Data Set of two, and GH_996_1 runs the
nine cases of #996 over the Data Set of that issue.

GH_980_1 pins the analytic count over a partition that holds only a null, which reports 0. That
was already fixed by #990 and is not changed here: the fixture is what makes the count rework
answerable for it, since the two engines disagreed on it before (1 on pandas, 0 on DuckDb) and
the Data Set is the one attached to that issue.

---

Closes #959
Closes #996
Closes #980

[javihern98]

Looks good, thanks! 😊