Converting errors to transaction_op_error_context loses data for error validation

Background; FIT requires that the performer does error validation, which is different depending on if the error returned is a TransactionOperationFailed, or something else (e.g. DocAlreadyExists from ctx.insert()).

Currently in C++ every error returned at this stage is converted into a transaction_op_error_context, containing the error_code and optionally a key_value/query error_context, if the error is a non-transaction error.

Converting the error into a transaction_op_error_context results in some data being lost from the underlying error (e.g. rollback, retry & the to_raise). One approach would be to simply add the these fields into the transaction_op_error_context (perhaps as internal fields), however, the general SDK approach of having specific error_context types etc., and them being returned in each operation is currently being discussed, and may affect how this issue is solved.