fix(slo_corrections): strip null-valued mutually-exclusive fields before POST

[riyazsh]

Summary

Destination POST /api/v1/slo/correction rejects payloads that carry slo_id, slo_query, or end explicitly set to null. Error shape:

API input validation failed: {'slo_query': ['Field may not be null.']}

Source resources round-tripped from GET /api/v1/slo/correction/{id} carry both slo_id and slo_query with one of them explicitly null — they are mutually exclusive:

• Time-based correction: slo_id set, slo_query = null
• Query-based correction: slo_query set, slo_id = null

Similarly, recurring rrule corrections carry end = null (no fixed end time).

Change

Add attributes.slo_id, attributes.slo_query, and attributes.end to SLOCorrections.resource_config.non_nullable_attr. The existing del_null_attr helper (datadog_sync/utils/resource_utils.py:277) strips fields listed there when their value is None, via prep_resource. Same pattern already used for attributes.duration and attributes.rrule (mutually-exclusive with each other), and for downtimes.recurrence.until_date vs .until_occurrences.

Safety

• Symmetric strip: resources_handler.py:503 (resource) + :506 (destination_copy) both pass through prep_resource before check_diff — no spurious diff on state-vs-resource comparison when one side has null and the other has the field stripped.
• Order with connect_resources: connect_resources runs at resources_handler.py:496, before prep_resource at :503. Populated slo_id gets remapped first; only the null slo_id (the mutually-exclusive case) is stripped.
• is None semantics: del_null_attr at resource_utils.py:285 uses resource[k_list[0]] is None, so integer 0 and empty string are preserved (matches destination API semantics, since 0 is a valid unix timestamp for end).

Tests

New file tests/unit/test_slo_corrections.py with 6 unit tests:

1. test_time_based_correction_strips_null_slo_query — populated slo_id, null slo_query → slo_query stripped, slo_id + end preserved
2. test_query_based_correction_strips_null_slo_id_and_null_end — populated slo_query, null slo_id + null end → both stripped
3. test_populated_fields_are_preserved — nothing null → nothing stripped
4. test_zero_end_is_preserved_not_stripped — end: 0 (integer zero) is NOT null and must survive
5. test_check_diff_symmetric_strip_no_spurious_diff — symmetric prep_resource on both sides of check_diff yields empty diff; regression pin against a future asymmetric-prep refactor
6. test_null_duration_and_rrule_still_stripped_regression — existing duration + rrule behavior preserved

Local pytest run: 6/6 pass. Broader unit suite: 1067 pass, only pre-existing unrelated failures in test_custom_client_trust_env.py (present on unmodified main).

Rollback

Three-line revert of the non_nullable_attr list additions.

[riyazsh]

The Ensure labels check is failing because a changelog/* label is required. Requesting a maintainer to apply changelog/Fixed (or an equivalent — this is a defect fix, no new API or removal). I do not have label permissions on this repo via either my personal or EMU account.