fix(attribute-values): Properly support boolean attributes

[wmak]

Summary

The TraceItemAttributeValues RPC enumerates the distinct values an attribute
takes (used to populate filter-value autocomplete). It previously assumed every
attribute was a string, so querying a boolean attribute broke. This makes
the endpoint properly support booleans.

Changes

• Existence check on the right column. Booleans live in attributes_bool,
  not attributes_string. A small _ATTRIBUTE_TYPE_TO_COLUMN map drives the
  has(...) check by attribute type; unsupported types raise a clear
  BadSnubaRPCRequestException. Adding more types later is a one-line map entry.
• Substring match guarded to strings. value_substring_match compiles to a
  SQL LIKE, which only makes sense for strings — it now raises on non-string
  keys instead of producing a broken query.
• Lowercase boolean values. The response proto's values field is
  repeated string, so booleans must be serialized as strings. They're returned
  as lowercase "true"/"false" — matching how booleans are stringified across
  the EAP filter API — so the returned values round-trip as filter inputs.
  Serializing via bool(value) also guards against ClickHouse returning 0/1.

Tests

• test_boolean_case — enumerating a boolean attribute returns
  ["true", "false"] with the expected counts.
• test_substring_match_on_boolean_rejected — substring match on a boolean is
  rejected.
• test_unsupported_attribute_type_rejected — non-string/bool types are
  rejected.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 5928272. Configure here.}

[phacops]

Re: the Sentry review-bot findings about f.has() on the attributes_bool Map column (HIGH + MEDIUM) — I verified these and they're false positives: has(attributes_bool, key) does check key existence correctly in ClickHouse here, and missing-key rows are not miscounted as false.

To prove it I added test_boolean_existence_check, which sets a custom_flag bool on only 2 of 9 seeded items (one true, one false) and leaves it off the rest. With the original f.has() code the test passed with counts == [1, 1] — i.e. the 7 rows without the key were correctly excluded, not folded into false. (The pre-existing test_boolean_case couldn't catch this either way, since gen_item_message injects sentry.is_segment=True on every item.)

That said, I still switched the existence check from f.has() to f.mapContains() (ead4824): it's the correct/conventional function for Map columns used everywhere else in the RPC code, HashBucketFunctionTransformer rewrites it for the bucketed attributes_string/attributes_float maps as well as the un-bucketed attributes_bool, and it keeps these warnings from recurring. The new test now gives real coverage of the existence filter for both has and mapContains.