fix(drive): correct S3 presign region, push download status over socket

[safwansamsudeen]

Summary

• generate_presigned_url signs offline and never gets boto3's automatic region-redirect retry that a real S3 call gets. Signed against the client's default region, S3 rejected the URL outright with PermanentRedirect for any bucket outside that region — hit in prod, since drive.frappe.cloud lives in ap-south-1 and the client defaulted to us-east-1. Each bucket's real region is now detected via get_bucket_location (itself region-agnostic) and cached, and presigning uses a client that matches.
• Replaces the 2s download_status poll with a single socket subscription: build_download_archive publishes the terminal ready/failed state to the requesting user's realtime room, and the client just listens (with one immediate status check as a race guard for a build finishing before the listener attaches, plus the existing timeout as a fallback if the event is ever dropped).
• Swaps the manual "preparing" toast for frappe-ui's toast.promise — one loading toast that morphs into success/error in place instead of hand-tracking an id across separate toast() calls.

Builds on #359 (already merged) — this branch is off develop.

Test plan

• Verified the region bug and fix live against prod's ap-south-1 bucket via bench console (get_bucket_location + region-scoped presign)
• Download a folder locally end-to-end: confirm the "Preparing download…" toast appears once, updates in place to "Download prepared", and the archive downloads
• Force a build failure (e.g. delete the source file mid-build) and confirm the toast shows the error instead of hanging
• Confirm no more download_status polling in the network tab — only one socket event drives completion

🤖 Generated with Claude Code