Debug TypeError in ethnicity measures for post-COVID cohorts

Thanks to @iaindillingham for helping to identify the issue https://github.com/opensafely-core/ehrql/issues/2445 and issue addressed by the PR: https://github.com/opensafely/WinterPressuresDescriptive/pull/10

The error was due to how the variable exp_bin_eth_missing was defined. Specifically, it used .is_null() on a Series derived from .to_category():

exp_bin_eth_missing = tmp_exp_cat_ethnicity.is_null()

However, .to_category() returns a Series that wraps a Case, and by default assigns None to unmatched codes. Unfortunately, .is_null() does not correctly detect these None values, so the resulting Boolean Series included unexpected Nones. This caused ehrQL to raise a TypeError when trying to count True values in the measure numerator.

The fix was to explicitly specify a default value in .to_category():

.to_category(ethnicity_snomed, default="Missing")

This ensures that unmatched codes are mapped to a string value ("Missing") rather than None, allowing missing ethnicity to be safely identified using:

exp_bin_eth_missing = tmp_exp_cat_ethnicity == "Missing"

This resolved the issue by producing a clean Boolean Series with only True/False values, avoiding the use of None in downstream measure calculations.

Review ehrql issue 2445 and WinterPressuresDescriptive PR 10 first, then trace the ethnicity measure around exp_bin_eth_missing and tmp_exp_cat_ethnicity. The issue is resolved when the measure no longer raises a TypeError while counting missing-ethnicity values and produces only boolean values for that condition.