Wake backoff deduplication

Source: tests/bdd/features/watcher_backoff.feature

Feature: Wake backoff deduplication
  The watcher tracks per-(bead, agent) wake history to prevent repeated
  wakes when an agent fails to claim a bead within a cron cycle.

  @bdd
  Scenario: First wake is allowed
    Given a watcher state with no prior wakes
    When check_wake_backoff is called for bead "ic-1" and agent "atlas"
    Then the result should be True

  @bdd
  Scenario: Within-backoff wake is skipped
    Given a watcher state with no prior wakes
    And a wake was recorded for bead "ic-1" and agent "atlas" at time 1000000.0
    When check_wake_backoff is called for bead "ic-1" and agent "atlas" at time 1000001.0
    Then the result should be False

  @bdd
  Scenario: Post-backoff wake is allowed
    Given a watcher state with no prior wakes
    And a wake was recorded for bead "ic-1" and agent "atlas" at time 1000000.0
    When check_wake_backoff is called for bead "ic-1" and agent "atlas" at time 1000901.0
    Then the result should be True

  @bdd
  Scenario: Max attempts stops all wakes
    Given a watcher state with no prior wakes
    And 3 wakes were recorded for bead "ic-1" and agent "atlas" long ago
    When check_wake_backoff is called for bead "ic-1" and agent "atlas" at time 9999999.0
    Then the result should be False

  @bdd
  Scenario: GC removes beads that left ready status
    Given a watcher state with no prior wakes
    And a wake was recorded for bead "ic-1" and agent "atlas" at time 1000000.0
    And a wake was recorded for bead "ic-2" and agent "atlas" at time 1000000.0
    When gc_wake_backoff is called with ready beads {"ic-2"}
    Then bead "ic-1" entries are removed from wake_backoff
    And bead "ic-2" entries remain in wake_backoff