fix(memory): skip hindsight daemon startup when running as root

[LehaoLin]

What does this PR do?

When hermes runs as root with local_embedded memory mode, the embedded PostgreSQL daemon (hindsight-api --daemon) cannot start because initdb refuses to run as root. The daemon then retries endlessly (~3 minutes per cycle), consuming CPU and memory in a tight loop. The only visible symptom is hermes hanging on startup with no clear error message.

This patch adds an early os.geteuid() == 0 check in the daemon startup path. When running as root, it skips the daemon entirely and logs a clear warning message telling the user to either:

• Run hermes as a non-root user, or
• Switch to cloud or local_external mode via hermes memory setup

Related Issue

Distinct from #13101 (which addresses missing hindsight-all package). This PR addresses the root-user daemon hang.

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Changes Made

• plugins/memory/hindsight/__init__.py: Added os.geteuid() check before daemon startup in initialize(). When root is detected, logs warning and skips daemon instead of entering an infinite retry loop.

How to Test

1. Run hermes as root with hindsight configured in local_embedded mode
2. Without this fix: hermes hangs on startup, hindsight-api --daemon process spins at ~30% CPU with log showing initdb: error: cannot be run as root repeating every ~3 minutes
3. With this fix: hermes starts normally, a single warning is logged explaining the issue and suggesting alternatives
4. Verify that running as a non-root user with local_embedded still works (no regression)

Checklist

• My commit messages follow Conventional Commits
• My PR contains only changes related to this fix
• Cross-platform: uses os.geteuid() (POSIX) with hasattr guard

[teknium1]

Salvaged and merged via #50308 (commit 7bc6f18 on main) — your authorship is preserved in git log.

Your branch had gone stale: it wrapped the pre-refactor _start_daemon body (hardcoded ~/.hindsight/profiles/*.env + manual env parsing), which has since been extracted into helper functions, so the diff couldn't be cherry-picked cleanly without reverting that refactor. I reapplied your substantive contribution — the os.geteuid() == 0 root guard — onto current main, plus added a stderr warning so the failure is user-visible (the issue's "zero feedback" complaint) and a unit test covering the root + non-root paths. Thanks for the fix!