feat: 轻量拓展分析 + WGS2GWAS 阶段支持 + 结果区域重构 + 物种角色解析增强 + 会话 Job 隔离

【轻量拓展分析】
- 新增 gfa_core/extension_stage_lightweight/ 核心模块，支持用户上传代码/数据/原始命令
- 前端卡片式交互流程：候选核对 → 验证 → 结果审查 → 授权发布 → 运行
- Celery 不受限任务 + 轮询机制 + sessionStorage 操作恢复
- 启动脚本新增 BBAV5_LIGHTWEIGHT_* 环境变量

【WGS2GWAS 支持】
- 新增 13 个 WGS 阶段定义（QC → MAF/Missing Filter）
- email_haplotype_region.py 新增 LTR-RT 分析功能

【结果区域重构】
- 单层折叠 → 三组双层折叠：RNA-seq / 基因家族 / WGS2GWAS
- 结果按执行时间排序，保留最新尝试

【物种角色解析增强】
- 新增 _resolve_subfamily_request_species_roles()
- 支持 reference 角色的识别
- 直接 stage 请求保留 intent 中的物种角色

【会话 Job 隔离】
- rebind_job() 跨 job 时创建新会话，返回新 conversation_id

【前端交互增强】
- RAG 文献问答模式
- NL 触发 config/samples 自动填写、亚家族确认表、基因家族成员确认表
- 执行状态从后端 resource_tasks 恢复

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

tasks.py

@@ -13,6 +13,19 @@ sys.path.insert(0, str(PROJECT_ROOT))
 from celery_app import app
 
 
+@app.task(
+    bind=True,
+    name="tasks.execute_lightweight_extension_operation",
+    time_limit=None,
+    soft_time_limit=None,
+)
+def execute_lightweight_extension_operation(self, payload: dict) -> dict:
+    """Run one lightweight-extension operation without a Celery deadline."""
+    from gfa_core.extension_stage_lightweight.api import execute_queued_operation
+
+    return execute_queued_operation(payload)
+
+
 def normalize_stage_params(stage: str, params: dict | None) -> tuple[dict, dict]:
     """Clamp per-stage CPU at the worker boundary and return resource metadata."""
     params = dict(params or {})
@@ -27,16 +40,19 @@ def normalize_stage_params(stage: str, params: dict | None) -> tuple[dict, dict]
 
             explicit_cpu = requested_cpu not in {None, "", 0, "0"}
             project_limit = rnaseq_project_cpu_limit(requested_cpu if explicit_cpu else None)
+            explicit_threads = params.get("threads_per_sample") or params.get("threadsPerSample")
             plan = recommend_rnaseq_parallelism(
                 total_cpu=project_limit,
                 sample_parallel_jobs=params.get("sample_parallel_jobs") or params.get("sampleParallelJobs"),
-                threads_per_sample=params.get("threads_per_sample") or params.get("threadsPerSample"),
-                sample_count=params.get("sample_count") or params.get("sampleCount"),
                 stage=stage,
             )
-            effective_cpu = clamp_stage_cpu(stage, plan["total_cpu"])
+            effective_cpu = clamp_stage_cpu(stage, project_limit)
             params["sample_parallel_jobs"] = plan["sample_parallel_jobs"]
-            params["threads_per_sample"] = plan["threads_per_sample"]
+            if explicit_threads:
+                params["threads_per_sample"] = max(1, int(explicit_threads))
+            else:
+                params.pop("threads_per_sample", None)
+                params.pop("threadsPerSample", None)
             params["rnaseq_project_max_cpu"] = plan["project_cpu_limit"]
             params["resource_decision"] = {
                 "stage": stage,
@@ -44,8 +60,11 @@ def normalize_stage_params(stage: str, params: dict | None) -> tuple[dict, dict]
                 "effective_cpu": effective_cpu,
                 "project_cpu_limit": plan["project_cpu_limit"],
                 "sample_parallel_jobs": plan["sample_parallel_jobs"],
-                "threads_per_sample": plan["threads_per_sample"],
-                "reason": f"{stage} project CPU budget split across sample-level parallel jobs",
+                "threads_per_sample": max(1, int(explicit_threads)) if explicit_threads else None,
+                "reason": (
+                    f"{stage} preserves the selected total CPU budget; per-sample threads "
+                    "are derived after the real sample count is known"
+                ),
             }
     except Exception as exc:
         requested_cpu = params.get("requested_cpu", params.get("cpu"))