feat: UI revamp - sticky headers, tabs, failure reasons, notif filters

[miguel502]

Lock table headers while scrolling in both apps: DataTable opts in by default, and the hand-rolled tables (activity, chain overview, provider breakdown, workflows) get bounded scroll boxes. #317

Consolidate multi-table screens into tabs: middleman Suppliers (Suppliers/Activity/Overview) and provider Keys (Keys/Activity), with the active tab persisted in the URL. #317

Surface the failure reason on failed transactions in both apps’ tables, reading the log/message columns through a shared failureReasonDisplay helper. #317

Add server-side filtering to notification history (event type, read/unread, channel) in both apps; the unread badge count stays unfiltered. #317

Make the left sidebar collapsible to an icon rail with a header toggle, hidden on the landing/auth pages. #317

Add unit tests for the notification filter conditions (both apps) and the shared failure-reason helper. #317

[jorgecuesta]

One thing worth a look before merging, and it's minor:

In KeysTabs.tsx (provider) the refetchInterval: 4000 ended up ungated. Since the query is now always mounted (you lifted it above the tabs), the Keys page fires a server-action call every 4s permanently, whether or not there's pending activity, on any tab. That poll used to live inside ActivitiesSection, which only mounted on the Activity tab, so effectively it's a new always-on poll. In middleman you gated it nicely (pendingOperations > 0 ? 7000 : false) — would be consistent to do the same here:

refetchInterval: (q) => (Object.keys(q.state.data?.byKey ?? {}).length > 0 ? 4000 : false),

And a micro nit, take it or leave it: in SuppliersTabs.tsx the badge counts byOperator while the poll gate looks at pendingOperations — if those ever diverge you could see the badge showing N with the poll already stopped. It's a rare edge and it's the same behavior the old counter had, so not worried, just noting it.

Otherwise 👍 from my side.