Add a dbt-optimizer agent: scan dbt projects for fixable issues with cost/impact reporting

从 docs/internal/2026-08-12-dbt-optimization-taxonomy-research.md 开始，将其中的检测领域映射到 Optimize 工作流背后的计划中 dbt-optimizer agent。定义只读扫描、候选选择、修复应用、成本/影响报告和 PR 输出，同时保留已声明的 warehouse-write、equivalence 和 query-history 安全门槛。完成的标准是指定的 agent 覆盖约定的范围，且不削弱这些要求。

Feature

Add a new native primary agent, dbt-optimizer, that scans an existing dbt project for concrete, fixable issues and proposes targeted fixes with cost and impact reporting. This is the agent behind the planned Optimize workflow.

Input: an existing dbt project (connected repository and environment).
Output: candidates (detected issues), fixes for user-selected candidates, impact reports (batch effect incl. estimated cost change), and PRs for applied fixes.

Detection scope (from the dbt-optimization taxonomy research, docs/internal/2026-08-12-dbt-optimization-taxonomy-research.md):

• materialization & incremental processing (strategy-aware: merge / delete+insert / insert_overwrite / microbatch, precondition-gated)
• warehouse physical design (clustering / partitioning / sort keys, query-history gated)
• SQL anti-patterns, DAG economics (fan-out, duplicate scans, dead models), run-level orchestration (Slim CI, threads, full-refresh overuse), tests/docs/storage

Safety requirements: scan phase is read-only; warehouse writes denied non-overridably; SQL rewrites gated on decidable equivalence; cost claims only from attributed query-history evidence.

Named dbt-optimizer (not optimizer) because more optimizer agents are planned later.