[jasperblues]

Ports drivine4j's #495 (liberation-data/drivine4j#498) and #497 (liberation-data/drivine4j#501), with liberation-data/drivine-cypher-spec#40 and #41. The compiler follows Drivine for TypeScript's port of the same two.

Calendar fields (#312)

• date, localdatetime and datetime take {year, week[, dayOfWeek]}, {year, ordinalDay} and {year, quarter[, dayOfQuarter]}, with a time and a timezone beside them. localtime and time refuse these fields.
• Where the year and the fields are written as numbers, the date is worked out when the statement is compiled and made as {year, month, day} is. That gives Neo4j's refusals at compile time: fields of two calendars, a day without its week or quarter, no year, a number past its range or not whole, and a year past ±999,999,999. Week 53 of a 52-week year and a day of the quarter up to 92 run on into the next. Ordinal day 366 of a common year fails.
• Where an expression gives a field, Temporals.of_calendar works the date out in SQL. A field past its range gives day 0, and the store fails on it. A moment's date is read as a date first, because Oracle's JSON_VALUE reads a moment whose day is -00 as null.
• dayOfQuarter can now be read off a temporal value.

HAVING (#313)

• In a WITH that aggregates, or is DISTINCT, the WHERE filters the groups. An aggregate there is now allowed, except inside a comprehension's, a quantifier's, a reduce's or a pattern's own expressions. The refusal's message now reads "… unless its WITH aggregates, …".
• _having makes each aggregate in that WHERE an item of the WITH:
  • Each new item is named #havingN by _hidden, which also names _anew_variable's variables, so the name is no text of the statement. Where an item already is that aggregate, its name is read instead.
  • A key written as its item is, before its name, is read as that item.
  • Keys and aggregates compare as grouping compares them. A function's name matches in any case, a property's name only as written, and count(DISTINCT i) is not count(i).
  • The made items aren't bound after the WITH.
• The WHERE is still met after ORDER BY, SKIP and LIMIT. A CALL that imports the row reads it after the WITH too.

Spec at v0.19.0. HELD (tests/cypher/cypher_spec.py) holds the steps a later port owns. Each held step must still fail; one that passes fails the test, so its hold comes out when its port lands:

• Compiled Cypher: a null in a map and a missing key group, de-duplicate and sort apart on the SQL stores (port of drivine4j#500) #314: "A null in a map and a missing key are one null" (Postgres steps 1–23, 25, 26; Oracle steps 1–27, 29) and "A null in a written row and a missing key are one null".
• Compiled Cypher: a grouping key read in an order term's body fails, and one in any(y IN … WHERE y = e) groups by its variable (port of drivine4j#505) #316: "Implicit grouping key" steps 29–31, 33–35, 37–40 and 42–44.
• Cypher on SQL stores: make a date, a time or a moment from another temporal value, as {date: d, week: 2} or {datetime: dt, timezone: '+05:00'} #317: the temporal-from-another cases in temporal.yaml and refusals.yaml, and steps 1 and 4 of "Temporal map not compiled yet".

The golden statements are unchanged. drivine4j's CompilerRefusalTest, CypherQueryContract and CypherTemporalContract cases that the spec doesn't hold are ported, and the sync twins are regenerated.

Fixes #312
Fixes #313