c/driver_manager: Python bindings can segfault if AdbcDatabaseInit() errors

[paleolimbot]

What happened?

When testing the framework driver in #2186, the driver manager segfaulted when the database init errored. The segfault happened here:

arrow-adbc/c/driver_manager/adbc_driver_manager.cc

Line 670 in 5248c34

return error->private_driver->ErrorGetDetailCount(error);

...because the function pointer for the error detail was nullptr.

I think this is because the driver had been released on a call to SetOption():

arrow-adbc/c/driver_manager/adbc_driver_manager.cc

Line 956 in 5248c34

database->private_driver->release(database->private_driver, error);

I am wondering if we hadn't seen this because typically database options are checked when the connection is initialized (or because drivers weren't zeroing the function pointers on release until using the framework).

An easy stopgap would be to also check that error->private_driver->ErrorGetDetail != nullptr here:

arrow-adbc/c/driver_manager/adbc_driver_manager.cc

Lines 667 to 673 in 5248c34

	int AdbcErrorGetDetailCount(const struct AdbcError* error) {
	if (error->vendor_code == ADBC_ERROR_VENDOR_CODE_PRIVATE_DATA && error->private_data &&
	error->private_driver) {
	return error->private_driver->ErrorGetDetailCount(error);
	}
	return 0;
	}

..or to zero out the error->private_driver after releasing the driver.

Stack Trace

No response

How can we reproduce the bug?

No response

Environment/Setup

No response

[lidavidm]

I think we should make both checks

This sort of thing is also why unloading drivers in the driver manager is hard/has been punted on: errors stick around and need a way to be released...