Allow ParquetReader to read LARGE_STRING, LARGE_BINARY, LARGE_LIST types from Arrow

[arhamchopra]

This PR adds support for reading LARGE_STRING, LARGE_BINARY, LARGE_LIST Arrow types in the ParquetReader
Resolves #398

With this change, the ParquetReader can be used with Arrow Large Types especially in polars more seamlessly:

import polars as pl
from datetime import datetime, timedelta
# Write polars DF to parquet
df = pl.DataFrame({
    "dt": [datetime.now() + timedelta(seconds=i) for i in range(10)],
    "str": ["a" for i in range(10)],                         # polars uses large_str for strings
    "list_of_str": [["a"] for i in range(10)],               # polars uses large_list for lists
    "binary": [b"a" for i in range(10)],                     # polars uses large_binary for binary data
    "list_of_binary": [[b"a"] for i in range(10)],
})

# df.to_arrow().schema
# dt: timestamp[us]
# str: large_string
# list_of_str: large_list<item: large_string>
#   child 0, item: large_string
# binary: large_binary
# list_of_binary: large_list<item: large_binary>
#   child 0, item: large_binary

import csp
from csp.adapters.parquet import ParquetReader
class MyStruct(csp.Struct):
    dt: datetime
    str: str
    list_of_str: csp.typing.Numpy1DArray[str]
    binary: str
    list_of_binary: csp.typing.Numpy1DArray[str]

@csp.graph
def my_graph() -> csp.ts[MyStruct]:
    reader = ParquetReader([df.to_arrow()], time_column="dt", read_from_memory_tables=True)
    return reader.subscribe_all(MyStruct, MyStruct.default_field_map())

[AdamGlustein]

Approving but would appreciate if @timkpaine or @robambalu can take a final look as well, as they know the Arrow API better than myself.