feat(experiments): refine launchdarkly-experiment-setup skill for new MCP tool shapes

[hsadhvani]

Summary

Closes EXPT-4361. Updates the experiment setup skill so the JSON payloads it teaches an agent to send match the actual LaunchDarkly MCP tool shapes (the LD REST API shapes used by gram-functions):

• create-experiment now uses the required nested iteration object, with hypothesis, metrics, treatments (each with allocationPercent + a parameters array), a flags map keyed by flagKey with ruleId/flagConfigVersion, and primarySingleMetricKey. Documents optional top-level tags, methodology, dataSource, analysisConfig, plus iteration-level attributes, covariateId, canReshuffleTraffic.
• update-experiment is described as mutableFieldsByStatus-aware: what's editable depends on iteration status, and skipped fields come back in the response.
• Adds workflow steps for the new tools: save-and-start-experiment-iteration (mid-experiment design changes in one call) and stop-experiment-iteration (with winningTreatmentId / winningReason).
• Drops the bogus reshuffle: true from the start step; mentions changeJustification instead.
• Bumps skill version to 0.2.0 and adds a CHANGELOG entry.

Testing approaches

• Static check against gram-functions tool inputSchemas — every JSON example field name, nesting, and type matches src/tools/experiments.ts in launchdarkly/gram-functions.
• Recommended follow-up: install the skill locally as a Claude Code plugin and walk through create → start → save-and-start → stop against a test LD project once gram-functions is reachable from your MCP config.

Feature flags

N/A

[github-actions]

Skill eval results

Skill	Before	After	Δ
agentcontrol/configs-create	100/100 (4/4)	75/100 (3/4)	-25
agentcontrol/configs-update	80/100 (4/5)	80/100 (4/5)	no change
agentcontrol/configs-variations	80/100 (4/5)	80/100 (4/5)	no change
agentcontrol/tools	75/100 (3/4)	75/100 (3/4)	no change
feature-flags/launchdarkly-flag-create	100/100 (3/3)	100/100 (3/3)	no change

Only suites whose source actually changed since their last recorded score were re-run. Soft-failing while we stabilise the baseline.