Fix #758: Clear ARROW_FLAG_NULLABLE on the entries struct of MAP columns

[adsharma]

Fixes #758.

Problem

Any query returning a `MAP` column is unreadable through the Arrow result
API. Consumers (e.g. arrow-java's `MapVector`) reject the schema Ladybug
produces with:

```
IllegalArgumentException: Map data should be a non-nullable struct type
```

This also breaks `RETURN n.*` and `RETURN n`, since including a single
`MAP` column makes the whole Arrow read path unusable for that table.

Root cause

`ArrowConverter::setArrowFormat`, `LogicalTypeID::MAP` case
(`src/common/arrow/arrow_converter.cpp`), only cleared
`ARROW_FLAG_NULLABLE` on the key child of the entries struct. The Arrow
columnar map layout requires both the entries struct and the key child
to be non-nullable:

The entries struct itself must be non-nullable, as is its key child.
— https://arrow.apache.org/docs/format/Columnar.html#map-layout

The fix is a single added line, applied immediately before the existing
key-child clear:

```cpp
child.children[0]->flags &=
~ARROW_FLAG_NULLABLE; // Map's entries struct must be non-nullable
```

Test

Added `ArrowTest.mapColumnArrowSchemaHasNonNullableEntriesAndKey` in
`test/api/arrow_test.cpp`. The test creates a `MAP(STRING, STRING)` column,
exports the Arrow schema through `getArrowSchema()`, and asserts:

• the map column has format `+m` and is itself nullable,
• the entries struct is named `entries` and is non-nullable,
• the key child is non-nullable, and
• the value child is nullable (values follow the value type's nullability).

Notes

• The C data interface does not treat child names as normative, so the
  upper-case `KEY`/`VALUE` naming mentioned in the issue is left for a
  separate change.
• A Java round trip with a locally patched `liblbug` cannot be verified
  end to end because the released `com.ladybugdb:lbug` artifact statically
  links its own engine build, but the flag is what `arrow-java`'s exception
  names, and it is now correct.