Sets a periodic AI credit limit for an End User. The limit rule is stored in Redis;
actual consumption is sourced from ss_ai_message. After saving the rule, the endpoint
immediately aggregates the database consumption within the new period and returns the
quota status.

Semantics:

• The period is a left-closed, right-open interval [startTime, endTime), i.e.
  message_time >= startTime AND message_time < endTime.
• maximumLimitCredits must be a JSON integer >= 0. Decimals and strings are rejected.
  When it is 0, the End User is not allowed to use AI regardless of historical usage.
• One End User keeps only one active rule. A subsequent PUT overwrites the previous rule
  and recomputes usage against the new period.
• The rule is scoped by applicationId + endUserId and is independent of login tokens;
  multiple sessions share the same rule and the same aggregated consumption.
• usedCredits is ROUND(SUM(ss_ai_message.total_credits)) within the period. All quota
  fields are integers.

Enforcement:

• The End User quota enforcement is gated by the AI_END_USER_QUOTA capability. When the
  capability is not enabled, AI calls only check the Application shared pool and the
  End User quota rule is ignored.
• When enforcement is enabled, the quota rule is checked on every AI call. A missing,
  not-yet-started, expired, or used-up rule blocks the AI call and returns the
  corresponding endUserAiQuota* error below. Account users are not affected by End
  User quota and never receive these errors.
• The endUserAiQuota* errors are returned only when an AI call is made under an
  End User token after enforcement is enabled; the PUT endpoint itself does not return
  them for rule validation failures (those return invalidParameter or
  aiQuotaEvaluationFailed).