feat(gooddata-eval): add KDA-skill agentic evaluator

Adds kda_skill.py to gooddata-eval, evaluating the chatbot's
create_key_driver_analysis/execute_key_driver_analysis tool calls against the
agent_kda_skill Langfuse dataset. Scope is completion (kda_triggered ->
executed -> success -> turn_completed), not per-field correctness -- per-field
checks are computed and logged as kda_-prefixed informational scores for a
follow-up ticket (QA-28699), and are None (not False) both when
expected_output has no key for a field AND when that field's own
precondition (kda_triggered, or executed+success for Summary) isn't met.

Latency is measured directly by the harness, not re-derived from Langfuse
after the fact: ChatResult.turn_wall_clock_sec is set by ChatClient around
the whole request (from just before opening the stream, so it includes
connection/server setup time the caller actually waits through, excluding
only a transient retry's own backoff sleep), stamped on both the successful
result and any ChatError.partial_result so a turn that dies mid-stream after
KDA already succeeded still gets a number. KdaRunResult.turn_wall_clock_sec
captures it in _run_once at the exact point create_args gets set -- atomic
with the kda_triggered signal, not a separate step that can fail
independently. Logged as the kda_turn_wall_clock_sec Langfuse score;
combo_report.py (gdc-nas) reads it directly, with no trace re-querying or
observation-shape matching needed on that side anymore.

Adds ChatResult.stream_ended (from the SSE response_ended event, derived --
not a raw server field, see sse_client.py's _RESPONSE_ENDED_EVENT comment).
turn_completed requires both stream_ended AND a non-empty text_response -- a
stream that ends cleanly but delivers nothing to the user isn't a completed
turn either, and a turn cut off mid-answer can still emit partial, non-empty
text before dying.

Fixes from review:
- kda_ prefix pass_at_k/pass_power_k Langfuse scores -- unprefixed, "pass_at_2"
  at k=2 collides with visualization.py's own score name that gdc-nas's
  combo_report.py.verdict() checks first, silently misfiling every KDA record
  as visualization once KDA_RUN_K=2 is ever set.
- Relative (not absolute) tolerance for Summary's revenue-scale values by
  default; change is checked against reference_value's scale, not against
  itself. Also recognizes an explicit absolute_tolerance key -- every real
  agent_kda_skill dataset item uses it, which this module never read before,
  silently running ~3000x looser than the dataset author intended. Warns on
  any other *tolerance* key so a future typo surfaces instead of repeating.
- Filters compared as canonicalized sets, not order-sensitive lists.
- except Exception (not except ChatError) around send_message -- a stream cut
  off mid-turn raises a raw httpx transport error, which a narrower catch
  would let escape uncaught, skipping Langfuse scoring entirely for that run.
- ChatError/TransientChatError carry partial_result so tool calls that
  already succeeded before a later, unrelated stream error aren't discarded
  and misreported as "the agent never called KDA at all".
- turn_completed resets to False on an exception, so a crash on a later
  disambiguation iteration can't leave a stale True from an earlier iteration.
- kda_disambiguated logged and immediately nulls the six *_correct
  informational fields: the simulated user reply names the acceptable
  candidate(s) drawn from expected_output itself, so those fields aren't an
  independent signal once a run went through disambiguation.
- KDA-specific _is_asking_clarification instead of a heuristic shared with
  metric_skill.py/conversation.py, which had silently drifted apart; strips a
  leading "to clarify, " discourse marker before its substring checks, since
  that phrase means "in other words" in a final answer, not a request for one.
- run_agentic_kda_skill rejects k < 1 -- the single initial run happens
  unconditionally regardless of k, so a bad env-driven KDA_RUN_K value (0, a
  typo, negative) previously ran silently once instead of surfacing the bad
  config, indistinguishable from a deliberate k=1.
- Regression tests for find_traces_per_conversation's None-safety and
  _filters_match's isinstance guard.
- _MAX_LATENCY_SEC (this module's own hardcoded 60s, used only for
  value_score's speed component) now cross-references gdc-nas's independently
  configured _KDA_MAX_TURN_SEC in a comment -- the two aren't shared code and
  nothing enforces them staying in sync if one changes without the other.
- stream_ended is now set the moment the response_ended event: line itself is
  parsed, not its data: line -- an event with no data payload at all
  previously left the flag unset.

JIRA: QA-28800

Author: FrankHuynh

packages/gooddata-eval/src/gooddata_eval/core/agentic/__init__.py

@@ -30,6 +30,14 @@
     evaluate_agentic_guardrail,
     run_agentic_guardrail,
 )
+from gooddata_eval.core.agentic.kda_skill import (
+    AgenticKdaSummary,
+    KdaEvaluation,
+    KdaRunResult,
+    KdaSkillAssertionError,
+    evaluate_agentic_kda_skill,
+    run_agentic_kda_skill,
+)
 from gooddata_eval.core.agentic.metric_skill import (
     AgenticMetricSummary,
     MetricRunResult,
@@ -56,6 +64,7 @@
     "AgenticAlertSummary",
     "AgenticGeneralQuestionSummary",
     "AgenticGuardrailSummary",
+    "AgenticKdaSummary",
     "AgenticMetricSummary",
     "AgenticSearchSummary",
     "AgenticRunSummary",
@@ -69,6 +78,9 @@
     "GeneralQuestionResult",
     "GuardrailAssertionError",
     "GuardrailResult",
+    "KdaEvaluation",
+    "KdaRunResult",
+    "KdaSkillAssertionError",
     "MetricRunResult",
     "MetricSkillAssertionError",
     "RunResult",
@@ -81,13 +93,15 @@
     "evaluate_agentic_conversation",
     "evaluate_agentic_general_question",
     "evaluate_agentic_guardrail",
+    "evaluate_agentic_kda_skill",
     "evaluate_agentic_metric_skill",
     "evaluate_agentic_search_tool",
     "evaluate_agentic_visualization",
     "run_agentic_alert_skill",
     "run_agentic_conversation",
     "run_agentic_general_question",
     "run_agentic_guardrail",
+    "run_agentic_kda_skill",
     "run_agentic_metric_skill",
     "run_agentic_search_tool",
     "run_agentic_visualization",