Prevalence percentage filter not showing all values when set to 0%

Expected behavior

Setting prevalence filter to 0% should show all values.

Actual behavior

For example: DemographicsAgeGroup, certain age groups are being filtered from the result because they are a relatively small proportion of the population.

Steps to reproduce behavior

1. Create cohort with a few outlier age groups (either very young or very old
2. Run Cohort characterization with demographics age group
3. View results, note by default the 1% filter may omit the small age group
4. Change the filter to 0%, and the results still filter

Example result from the UI and from the raw query:

UI:

Raw Query:

Note, the last age group (65-69) is missing from the exported results (the export results matches the UI).

Query (SQL Server dialect):

-- DemographicsAgeGroup
-- analysis_id = 3

-- Feature construction
with cteCovariates (covariate_id, cohort_definition_id, sum_value) AS (
	SELECT CAST(FLOOR((YEAR(cohort_start_date) - year_of_birth) / 5) * 1000 + 3 AS BIGINT) AS covariate_id, -- analysis_id = 3

		cohort_definition_id,
		COUNT(*) AS sum_value
	FROM ohdsi_results.cohort cohort
	INNER JOIN dbo.person
		ON cohort.subject_id = person.person_id
	WHERE cohort.cohort_definition_id IN (13555)
	GROUP BY cohort_definition_id,
		FLOOR((YEAR(cohort_start_date) - year_of_birth) / 5)
), cteCovRef (covariate_id, covariate_name, analysis_id, concept_id) AS (
	SELECT covariate_id,
		CAST(CONCAT (
			'age group: ',
			RIGHT(CONCAT('   ', CAST(5 * (covariate_id - 3) / 1000 AS VARCHAR)), 3),
			' - ',
			RIGHT(CONCAT('   ', CAST((5 * (covariate_id - 3) / 1000) + 4 AS VARCHAR)), 3)
			) AS VARCHAR(512)) AS covariate_name,
		3 AS analysis_id,
		0 AS concept_id
	FROM (
		SELECT DISTINCT covariate_id
		FROM cteCovariates
	) t1
)
select c.cohort_definition_id, cr.covariate_name, c.sum_value
from cteCovRef cr
join cteCovariates c on cr.covariate_id = c.covariate_id
ORDER BY c.sum_value desc;

Start by reproducing the cohort characterization steps with a 0% prevalence filter and compare the UI and exported results with the raw SQL query shown in the issue. Trace where the prevalence threshold is applied, then verify that small groups such as age 65-69 appear in both UI and export results when the filter is 0%.