Markdown import reads every file with blocking std::fs::read_to_string inside async daemon tasks #2961

Problem

Running a large/slow markdown import — including the documented nodespace import dir ../nodespace-docs --auto-collection-routing workflow — can degrade concurrent daemon responsiveness (ai-chat streaming, WatchNodes delivery, other RPCs) for the duration of the import, because a blocking syscall run directly on a shared tokio worker thread prevents that worker from polling any other task until the read completes.

Root cause

import_single_file() (packages/daemon/src/services/import_service.rs:172) and the batch-import phase-1 loop in run_batch_import() (line 344) both call std::fs::read_to_string(path) directly inside async fns that run on the daemon's shared multi-threaded tokio runtime, with no spawn_blocking wrapper. Every other file-I/O call site in this crate (database_manager.rs, settings_service.rs) consistently uses tokio::fs, and this codebase has an established, explicitly-documented convention (assembly.rs, local_agent_service.rs) of wrapping slow synchronous work in spawn_blocking specifically to avoid parking a tokio worker thread — this file is the one outlier. run_batch_import's read loop is sequential, so the blocking reads happen one after another inside a single spawned task.

Suggested fix

Replace std::fs::read_to_string with tokio::fs::read_to_string (or wrap in spawn_blocking) in both import_single_file and run_batch_import's read loop, matching the crate's own established convention elsewhere.

Found via the 2026-09-25 full-repo code sweep (finding #11 of 84 in CODE_SWEEP_FINDINGS_2026-09-25.md). Severity: medium. Category: risky.