fix(lineage): hint column-level needs existing table; pin jsp node-key tags (#154)

[c2j]

Follow-up to #156 (merged as c88156c) — 落实对 PR #156 的两轮 review 中最后两项 low 级遗留（review 5130018719 后续复审发现）。

Closes #154 (补充)

变更

1. Missing 回退 note 补充列级前提（fix）

codeweb lineage "orders.customer_id"（orders 不存在）时，原输出为两段：

note: no table 'orders' found — interpreting 'orders.customer_id' as a table reference
No table found matching 'orders.customer_id'

用户看到的最终错误是拼接后的怪异表名，缺少修复指引。现 note 补充前提：

note: no table 'orders' found — interpreting 'orders.customer_id' as a table reference (for column-level lineage, the table must exist)

新增测试 should_hint_column_level_requires_table_when_target_missing 锁定（TDD Red→Green）。

2. roundtrip 测试钉住 jsp/jspsql 标签（test）

TYPE_TAG_PREFIXES 无条件包含 "jsp"/"jspsql"，但其 NodeKey 变体为 cfg(feature = "jsp")，should_detect_every_display_tag_roundtrip 从未覆盖它们（即使 --features jsp）。若 Display 格式变更，split_type_prefix 会静默失配且只表现为 table.column 误parse。

修法：测试内 #[cfg(feature = "jsp")] 块构造真实 JspPage/JspSql 变体并断言可检测；--features full 门禁下即被执行。常量文档同步说明。此为 characterization pin（当前行为已正确，无 Red）。

测试与门禁

• regress_issue_154_lineage_targets：8 passed（新增 1 个）
• should_detect_every_display_tag_roundtrip：默认 + --features jsp 双组合通过
• cargo fmt --all -- --check ✅ / cargo clippy --features full -- -D warnings ✅ / cargo test --features full -- --skip test_path_mapping_applied --skip test_serve_ ✅（基于最新 main，含 feat(serve): 调用链路面板展开/折叠 + has_more 截断标记 (#152) #155/fix(analyze): store 版本校验缺失导致升级二进制后 stale 缓存死循环 #157 合入后验证）

[c2j]

[suggestion] The new parenthetical (for column-level lineage, the table must exist) is appended on every TableLookup::Missing fallback, not only when the table is actually missing. That branch is also the successful schema.table recovery: codeweb lineage "bigfund.mid_yjqs_detail" first looks up table bigfund (a schema, so Missing), prints this note, then resolves the whole target as a table and succeeds. Users who typed a valid schema-qualified table now see a note implying column-level lineage failed because the table does not exist, even though stdout is a correct table-level result. should_fall_back_to_table_level_for_bare_schema_qualified_target still passes because it only checks for interpreting plus the original target. The new test covers only the dead-end missing_table.some_col case.

Suggestion: Keep the original Missing note for the schema.table recovery. Add the column-level prerequisite only when the subsequent table-level resolve also fails (the No table found matching path), or thread a “fell back from a column spec” flag and emit the extra sentence there. Extend the existing schema.table fallback test to assert the column-level parenthetical is absent on success.

[c2j]

[suggestion] The new JSP cases only assert split_type_prefix(key).is_some(). That does not pin the tags jsp / jspsql. JspSql Display (jspsql:file:line:hash) is one character away from JavaSql (javasql:file:line); if the JspSql arm were copy-pasted onto the javasql format, split_type_prefix would still succeed because javasql is already in TYPE_TAG_PREFIXES, and jspsql:… CLI keys would again last-dot-split. That is the drift this follow-up is meant to lock.

Suggestion: Assert the detected tag, e.g. assert_eq!(split_type_prefix(&jsp_page).map(|(tag, _)| tag), Some("jsp")) and Some("jspsql") for JspSql, rather than is_some().