[scheduler] Configurable outdated thresholds, fix early resource scrubbing

[cgalibern]

Problem

scrub_resources flags live resources as undef too early:

scrub_1m: scrub_resources: set 999 resmon status to undef (no live instance) for ...

Two causes:

• InstancePingFromNodeID returned early when no svcmon row needed a
  refresh, so resmon.updated and resmon_log_last.res_end were not
  refreshed. When svcmon was kept fresh by the daemon status feed,
  resmon.updated aged, and resources of live instances were scrubbed
  after 15 minutes.
• The age limits used by the scheduler tasks were hardcoded in SQL, so
  they could not be tuned to a deployment.

Changes

[worker] Fix resmon not refreshed on daemon ping

The svcmon, svcmon_log_last, resmon and resmon_log_last refreshes in
InstancePingFromNodeID now run independently.

[scheduler] Make task outdated thresholds configurable

• Each task that decides a value is outdated reads its age limit from
  scheduler.task.<task>.max_age. The defaults keep the previous
  hardcoded values, so behaviour is unchanged unless configured.
• The value accepts time.ParseDuration units plus d (15m, 25h,
  2d, 1d12h) and must be at least 1m.
• All scheduler.task.*.max_age keys are validated when the scheduler
  starts, so the scheduler and scheduler exec refuse to start with an
  invalid value.
• scrub_object no longer uses the v_outdated_services view; its
  query is now inlined with a configurable age. The database schema is
  unchanged and the view is left in place.

Key (scheduler.task.<key>.max_age)	Default
scrub_object, scrub_resources	15m
scrub_instances	21m
scrub_unfinished_actions	2h
scrub_resmon, scrub_pdf	1d
scrub_checks_live, scrub_diskinfo, scrub_svcdisks, scrub_stor_array	2d
scrub_node_hba, scrub_comp_status_unattached	7d
scrub_comp_status	31d
scrub_packages, scrub_patches	100d
scrub_static, scrub_tempviz	1h
alert_instances_not_updated	16m
alert_nodes_not_updated, alert_service_config_not_updated	25h
alert_checks_not_updated	1d
log_instances_not_updated	2h

Example:

scheduler:
  task:
    # max_age: the age after which the task considers a value outdated.
    # Units: s, m, h, d (e.g. 15m, 25h, 2d, 1d12h). Minimum 1m.
    # All max_age values are checked at startup; the scheduler refuses
    # to start if one is invalid.

    # alerts
    alert_checks_not_updated:
      max_age: 1d
    alert_instances_not_updated:
      max_age: 16m
    alert_nodes_not_updated:
      max_age: 25h
    alert_service_config_not_updated:
      max_age: 25h
    log_instances_not_updated:
      max_age: 2h

    # status scrubs (every minute)
    scrub_object:
      max_age: 15m
    scrub_resources:
      max_age: 15m
    scrub_instances:
      max_age: 21m

    # action scrubs (every 10 minutes)
    scrub_unfinished_actions:
      max_age: 2h

    # table purges (daily)
    scrub_checks_live:
      max_age: 2d
    scrub_comp_status:
      max_age: 31d
    scrub_comp_status_unattached:
      max_age: 7d
    scrub_diskinfo:
      max_age: 2d
    scrub_node_hba:
      max_age: 7d
    scrub_packages:
      max_age: 100d
    scrub_patches:
      max_age: 100d
    scrub_resmon:
      max_age: 1d
    scrub_stor_array:
      max_age: 2d
    scrub_svcdisks:
      max_age: 2d

    # file scrubs in directories.static
    scrub_static:
      max_age: 1h
    scrub_tempviz:
      max_age: 1h
    scrub_pdf:
      max_age: 1d

Operators changing these values should keep them consistent:
scrub_instances purges svcmon rows, so setting scrub_resources or
scrub_object above it means instances are purged before their
resources or objects would be flagged undef.