docs(audits): daemon thread-per-connection scalability ceiling (#1673)

[oferchen]

Summary

• Replace docs/audits/daemon-thread-per-connection.md with a focused, 116-line static audit covering the daemon accept loop and connection lifecycle.
• Cite the accept loop, worker spawn, and counter sites in crates/daemon/src/daemon/sections/server_runtime/, quantify per-session memory (BufferPool, ReorderBuffer, flist) and the ~30 k-thread Linux ceiling, and contrast oc-rsync's thread+catch_unwind model with upstream rsyncd's fork-per-conn process boundary.
• Document failure modes (SYN flood, slowloris on auth-stalled connections, lock contention on shared module state, O(n) reap latency) and mitigations: async tokio listener (Accept quoted fallback override values #1934/Normalize quoted fallback keywords #1935), daemon-level max connections enforcement on the existing ConnectionCounter, connection rate limit, bounded worker pool, reduced stack size.

Test plan

• Confirm the file is under 130 lines (wc -l).
• Confirm cited line numbers still resolve in accept_loop.rs, connection.rs, workers.rs, connection_counter.rs.
• CI: docs-only change, expect fmt/clippy/test workflows to pass without behavioural impact.