Add more granular lazy execution flags in query compiler

[sfc-gh-joshi]

Modin allows backends to opt-in to certain frontend error validation checks based on the query compiler's lazy_execution flag. Some methods (like astype) use this to avoid materializing column labels, while others (like drop) avoid materializing row labels or counts.

Certain backends may defer execution for row/column axes differently: for example, Snowpark pandas eagerly evaluates columns but leaves row labels/counts lazy. We should split up the lazy_execution flag to allow backends to make this distinction in a more granular manner (proposed by @devin-petersohn):

• lazy_row_labels for backends that defer computations of index
• lazy_row_count to defer computation of len(index)
• lazy_column_types to defer computation of dtypes
• lazy_column_labels to defer computation of columns
• lazy_column_count to defer computation of len(columns)