feat(soc-optimization-unified): add SOCFW Framework Health dashboard

[scottbrumley]

The framework-health half of the 15 Sep dashboard split. Lifecycle-outcome dashboards moved to the NIST IR packs in #1169-#1171; this is the other half and it ships in core.

Command errors first, because the data found a live problem nothing was surfacing.

What it shows

Measured on a reference tenant over 30 days:

command executions with has_error	3,830 of 5,771 (66%)
every one of them	action_status = integration_unavailable
minutes credited to commands that never ran	7,200

Nine widgets

• Command Error Rate (%), Failed Commands, Minutes Credited to Failed Commands — the headline three
• Failing Commands: Universal → Vendor — the pair that matters, not the verb alone. soc-enrich-endpoint appears twice against two different vendor commands
• Failure Reason and Command Outcomes by Status — why failures happen, and the shape of all outcomes including simulated, which is shadow mode working rather than a fault
• Failing Commands - Detail — drilldown table: verb, product, vendor command, failed vs total, error rate, affected cases. Every failing pair currently reads 100%, so this is unavailability rather than flakiness
• Command Failures by Day — a single bad day dominates the 30-day average; the rate needs reading beside the trend
• Framework Write Failures — from playbook_tasks: socfw-post-to-dataset failed 1,274 times, so the execution dataset is missing rows and every metric built on it understates

Build notes

Structure copied from a shipped dashboard rather than authored from scratch. Two install rules learned the hard way, both via 101704:

• a table widget takes no | view clause — every other type requires one
• there is no line chart type in the shipped vocabulary (single, column, pie, funnel, table only)

default_dashboard_id left absent, matching the Value Driver dashboards. Widget keys conform to xql_<13-digit epoch>, row ids to row-<digits>.

Verification

Iterated through the dashboards REST API (/public_api/v1/dashboards/{get,insert,delete}) rather than repeated pack installs, so no orphaned dashboards were created while building. Every query run against a reference tenant before it went in the file. check_contribution including the upload step, green.