v0.8.4

Changed

• Spec sync: gpt-5.4 → gpt-5.5 for Developer + Independent Reviewer model references — OpenAI released gpt-5.5 (and gpt-5.5-mini) since the v0.8.0 model-upgrade Wave (2026-04-18) wrote gpt-5.4 / gpt-5.4-mini into the spec, and the Codex CLI's [notice.model_migrations] mechanism plus user runtime config (~/.codex/config.toml model = "gpt-5.5") have already moved actual usage forward. This release brings the iSparto spec into alignment with what the Codex runtime is already serving. Replaced both lowercase gpt-5.4 / gpt-5.4-mini (model names in code-fenced contexts and prose) and uppercase GPT-5.4 (Independent Reviewer narrative descriptions of the cross-provider blind-review model) across nine active-scope files: CLAUDE.md, CLAUDE-TEMPLATE.md, agents/independent-reviewer.md, docs/concepts.md, docs/collaboration-mode.md, docs/configuration.md, docs/design-decisions.md, docs/roles.md, docs/workflow.md. Tier 4 historical artifacts (docs/session-log.md, docs/independent-review.md, docs/observation-period/wave0-de-audit.md, CHANGELOG.md previously-released entries, and historical Wave entries inside docs/plan.md) intentionally NOT modified — those entries are frozen historical records of what was true at write time. Two-tier mapping unchanged (implementation = full model at xhigh; QA / quick fix = mini variant at high). Cost unchanged (ChatGPT Plus covers both). Fast Mode (service_tier = "fast") still applies uniformly. No functional behavior change — Lead/Teammate Codex prompts already default to whatever Codex CLI dispatches; this commit just makes the docs say what is actually happening.