feat(extensions/nanoarrow_ipc): Add single-threaded stream reader

[paleolimbot]

Higher level runtimes may be able to use the ArrowIpcDecoder (or more than one) and handle IO/parallelization using tools that are difficult to provide from C; however, for testing we do need the ability to read streams in their entirety. This PR provides a tool to do that based on an arbitrary bytes input.

This reduces the overhead of coordinating the various steps required to decode a stream to:

struct ArrowIpcInputStream input;
ArrowIpcInputStreamInit<Buffer|File|Custom Implementation>(&input, ...);

struct ArrowArrayStream stream;
ArrowIpcArrayStreamReaderInit(&stream, &input, nullptr);

struct ArrowSchema schema;
stream.get_schema(&stream, &schema);

struct ArrowArray array;
while (1) {
  stream.get_next(&stream, &array);

  if (array.release != NULL) {
    array.release(&array);
  } else {
    break;
  }
}

schema.release(&schema);
stream.release(&stream);

There is also a utility to read an entire stream from file or stdin:

$ ./dump_stream big.arrows 
Read Schema <0.000122 seconds>
struct
  state: string
  geometry: geoarrow.polygon{list}
    rings: list
      vertices: geoarrow.point{fixed_size_list(2)}
        xy: double
Read 23548499 rows in 1 batch(es) <1.623532 seconds>