Every req takes rx-nostr's random rxReqId, so concurrent hooks can collide

`useReq()` never passes an `rxReqId`, so every req falls back to rx-nostr's
default (`rx-req.ts:125, 220-222`):

```ts
rxReqId: params.rxReqId ?? getRandomDigitsString()
// `${Math.floor(Math.random() * 1_000_000)}`
```

That is a 10^6 space shared by every concurrent request in the page. Two hooks
that happen to draw the same number produce the same wire subId, and rx-nostr
then either drops one of them silently (backward: `connection.ts:181`) or lets
them overwrite each other and cross-talk (forward: `rx-nostr.ts:325`).

By the birthday bound the probability of at least one collision among `n`
concurrent reqs is roughly `1 - exp(-n^2 / 2e6)`: about 0.5% at 100, 12% at 500,
39% at 1000. A feed that renders a `` and a `` per item reaches
those counts on a long timeline — `src/routes/timeline/+page.svelte` and
`src/routes/n-plus-1/+page.svelte` both fan out this way.

The symptom when it happens is a hook that stays `'loading'` forever, or one
that receives another hook's events, with nothing in the logs.

Determined by code reading; not reproduced (it is probabilistic by nature).
The id in a captured wire log from an unrelated test — `["REQ","408976:0",...]`
— shows the default is in use.

## Direction

Pass an explicit `rxReqId` from a per-context counter. This removes the failure
mode entirely rather than making it rarer, and it costs nothing: all three
factories already accept one (`rx-req.ts:164`, `:188`, `:204-207`).

Related: #61 and #82 both concern the same req being reused; this one is about
distinct reqs colliding.