Durable dashboard session store

# F3.2 — Durable dashboard session store (nr-launcher)

⚠️ Needs #8280 aka F3.1 first.

## Context

When a dashboard uses FlowFuse User auth, the "this user is logged in" record is kept in an in-process `MemoryStore` inside the instance's Node-RED process. That has two consequences:

- **Restart:** the process dies on any restart/redeploy/suspend-resume → the store is wiped → every viewer is logged out at once (a team-wide re-auth/Forbidden wave).
- **HA:** each replica has its own store → a session on one pod is invisible to the others, so viewers keep re-authing.

F3.1 aka #8280 fixed the *cookie* surviving a restart; F3.2 is the other half — making the session *data* durable so the cookie actually has something to resolve to. Both are needed before "sessions survive a restart" is true.

**Fixes for users:** dashboard sessions survive an instance restart/redeploy and work across HA replicas. This is the actual F3 payoff.

## Options

| Option | What | Effort | Risk | Covers |
|---|---|---|---|---|
| **D. Forge-backed HTTP store** | custom store → new forge endpoint | Med–High | Med — new endpoint, but matches how the launcher already persists everything (HTTP-to-forge), **no infra** | Restart + HA |
| A. File store on PVC | `session-file-store` on the existing `/data/storage` RWX volume | Med | Med — gated on `persistentStorage` (**default off**); file-locking on shared FS | Restart + HA (only when enabled) |
| B. Redis/Valkey | shared store | High | High — instances can't reach Redis today; needs new infra + dep + failure mode | Restart + HA |
| C. Ingress cookie affinity | sticky routing, no store change | Med | Med — infra, ingress-specific | HA only (**not** restart) |

**Lean:** D — it fits how the launcher already works (everything persists over HTTP to forge), needs no new infrastructure, and solves both restart and HA. A is the cheaper shortcut *if* `persistentStorage` is already on for the affected instances. Avoid B (infra cost) and C (doesn't fix restart).

Start in nr-launcher at the dashboard's in-process MemoryStore and trace how session reads and writes reach the Node-RED process. Review prerequisite issue #8280 and the existing forge HTTP persistence path; the preferred direction is a forge-backed HTTP store. Done means sessions survive restart or redeploy and resolve consistently across HA replicas, with coverage for both behaviors.

Written by the indexing model from the issue text.