fix: Intel Mac support, RAM pre-flight check, TLS revocation, Ollama readiness poll, and model pref update on re-install

[ViperNetwork]

Resolves #49, #46, #60, #61, #62

Summary

• Mac launcher only downloads Apple Silicon build (Intel Macs unsupported) #49 — Intel Mac support: start-mac.command now uses uname -m to
  select the correct AnythingLLM DMG (-Silicon vs plain) at first-run
  download time. Ollama's ollama-darwin.zip is already a universal binary
  and is unchanged.

• Llama runner process has terminated with exit code 2 when loading large models (NemoMix/Qwen) #46 — RAM pre-flight check / OOM messaging: linux/preflight-check.sh
  gains a dedicated Step 4/5 that reads /proc/meminfo, grades total RAM
  against per-model thresholds (4/6/8 GB), and fails the check if < 4 GB.
  The three launchers (start-linux.sh, start-mac.command,
  install-core.ps1) now emit an advisory line when available RAM is below
  the threshold for the installed models. If Ollama exits before the
  readiness poll completes, the launchers now print an actionable OOM
  message instead of silently hanging.

• --ssl-no-revoke disables TLS certificate revocation on all binary downloads #60 — TLS revocation: Removed --ssl-no-revoke from all three
  curl.exe calls in install-core.ps1 (model files, Ollama binary,
  AnythingLLM installer). The flag disabled OCSP/CRL checking across all
  binary downloads, which is a meaningful security downgrade.

• Ollama server readiness check uses a fixed sleep instead of a health poll #61 — Ollama readiness poll: Replaced every fixed sleep/timeout
  with a curl/Invoke-WebRequest loop that polls GET /api/tags once per
  second, up to 30 s. Returns as soon as the server is ready; prints a
  warning on timeout. Affects start-linux.sh, start-mac.command,
  start-windows.bat, and install-core.ps1. (install-core.sh already
  had a correct poll loop.)

• Re-running installer to add a second model does not update OLLAMA_MODEL_PREF in .env #62 — OLLAMA_MODEL_PREF not updated on re-install: Both
  install-core.sh and install-core.ps1 now update only the
  OLLAMA_MODEL_PREF line in an existing .env via sed -i / regex
  replace. All other user settings (token limit, embedding engine, etc.) are
  preserved.

Breaking changes

None. All changes are additive or remove a security downgrade.

Notes for reviewers

• --ssl-no-revoke removal may surface failures in environments where OCSP
  endpoints are firewall-blocked. This is intentional; the correct fix there
  is to configure a trusted corporate CA, not to skip revocation globally.
• OLLAMA_MODEL_PREF is now always updated by the installer on re-runs.
  Users who manually changed the model in the AnythingLLM UI will have their
  preference overwritten the next time they run the installer. This is
  consistent with the documented behaviour ("re-run install.bat to add
  models") and matches what the first-time install already did.