Rolling window queries

CubeStore extension to implement rolling window measures in CubeJS.
Tests are in the CubeStore repository.

The syntax is of the form:
```
SELECT dim, key1, key2, other,
       ROLLING(SUM(x) RANGE BETWEEN 7 PRECEDING AND UNBOUNDED FOLLOWING)
       ROLLING(AVG(x) RANGE BETWEEN 7 PRECEDING AND UNBOUNDED FOLLOWING)
FROM input
ROLLING_WINDOW
  DIMENSION dim
  PARTITION BY key1, key2
  FROM 0 to 10 EVERY 2
```

Semantics are roughly:
  - compute rolling window aggregations over input data,
  - window "rolls over" the `DIMENSION` column, only values defined in
    the range by `FROM .. TO .. EVERY ..` are reported,
  - each "group" defined by columns in `PARTITION BY` is handled
    and reported independently,

Current limitations:
  - only integer ranges (timestamps and intervals are coming),
  - only ranges with up to 10M points are supported to avoid infinite
    loops are accidental DOS. This is still a fairly large limit and
    can lead to DOS given enough input data.

Author: ilya-biryukov

datafusion/Cargo.toml

@@ -48,7 +48,7 @@ ahash = "0.7"
 hashbrown = "0.11"
 arrow = { git = "https://github.com/cube-js/arrow-rs.git", branch = "cube", features = ["prettyprint"] }
 parquet = { git = "https://github.com/cube-js/arrow-rs.git", branch = "cube", features = ["arrow"] }
-sqlparser = "0.9.0"
+sqlparser = { git = "https://github.com/cube-js/sqlparser-rs.git", rev = "a04701767d9d6a9eb99cce9a64dee7e1b7c25b7c" }
 paste = "^1.0"
 num_cpus = "1.13.0"
 chrono = "0.4"

datafusion/src/cube_ext/mod.rs

@@ -18,6 +18,7 @@
 pub mod alias;
 pub mod join;
 pub mod joinagg;
+pub mod rolling;
 pub mod sequence;
 pub mod stream;
 pub mod util;