fix: SQLite 大批次查询与写入回滚 / SQLite chunked queries and rollback

[mfurkanakinci]

中文

大批次写入时，SQLite 的 IN 查询会超过 999 个变量的上限并导致整批失败。写入失败时也没有回滚，错误被宽泛地吞掉。

• 新增 trendradar/utils/sqlite.py，把 IN ({placeholders}) 按块执行（每块 500 个值）
• 热榜、RSS 和 MCP 解析里的大 IN 查询改走这条路径
• 写入失败时回滚连接，并以 StorageError 抛出不可忽略的存储错误
• 不引入 PostgreSQL 依赖

tests/test_sqlite_chunked_in.py 覆盖超过旧上限的 IN 查询，tests/test_sqlite_mixin.py 覆盖写入、合并、排名历史和 AI 标签生命周期。

English

Large batches can exceed SQLite's 999-variable limit on IN queries and crash the write. Failed writes were also left unrolled back, and the error was swallowed.

• Add trendradar/utils/sqlite.py and run IN ({placeholders}) in chunks of 500 values
• Use that path for the large IN queries in hotlist storage, RSS storage, and the MCP parser
• Roll back the connection when a write fails, and raise StorageError for storage failures that must not be ignored
• No PostgreSQL dependency

tests/test_sqlite_chunked_in.py covers an IN query past the old limit. tests/test_sqlite_mixin.py covers writes, merges, rank history, and the AI-tag lifecycle.