spin-node: honour multi-aggregator presets, enforce >=1 per subnet

[ch4r10t33r]

Summary

`spin-node.sh` enforced "exactly one aggregator per subnet": it kept only the first preset per subnet, reset every `isAggregator` flag to false on every deploy, and aborted with `expected exactly 1` if it ever saw more than one `true` row in a subnet. Operators who hand-edited `validator-config.yaml` for redundancy (e.g. `zeam_0` + `grandine_0` on subnet 0, `qlean_1` + `ethlambda_0` on subnet 1) silently lost their second aggregator on the next full deploy.

Behaviour change

Scenario	Before	After
Subnet has 0 presets	1 random pick	1 random pick (unchanged)
Subnet has 1 preset	preset honoured	preset honoured (unchanged)
Subnet has N>=2 presets	only first kept; rest reset to `false`	all N honoured
`--aggregator NODE` given	NODE replaces preset on its subnet	NODE replaces presets on its subnet (unchanged); other subnets honour their presets
Verification invariant	`agg_count == 1` per subnet	`agg_count >= 1` per subnet

The YAML mutation is now a single pass: each row is set to `true` iff its name is in the final aggregator set, else `false`. This replaces the previous reset-then-set sequence which left the file in a transiently empty state.

Why this is safe downstream

Every client cmd (`client-cmds/zeam-cmd.sh`, `ethlambda-cmd.sh`, `grandine-cmd.sh`, etc.) and every ansible role (`ansible/roles/*/tasks/main.yml`) reads `isAggregator` per node independently. Multiple `true` rows in the YAML naturally translate into multiple containers being launched with `--is-aggregator`. There is no global "single aggregator" assumption anywhere in the deploy path.

What stays the same

• `--restart-client` still skips the aggregator block entirely (no change to existing flags).
• The cross-subnet client-family uniqueness heuristic still applies to random fallbacks. Presets are exempt because the operator opted in.
• `--dry-run` still skips both YAML writes and the verify pass.

Test plan

Verified in a temp `NETWORK_DIR` with a 3-subnet, 8-node fixture:

• Subnet with two presets (`zeam_0` + `grandine_3`) → both kept.
• Subnet with one preset (`ethlambda_4`) → preset kept.
• Subnet with zero presets → one random pick, distinct family.
• `--aggregator ream_1` (subnet 1 = ream_1's subnet) → ream_1 replaces ethlambda_4 preset on subnet 1; subnet 0's two presets remain untouched.
• Reset all rows to `isAggregator: false` first, then run → exactly one random pick per subnet, distinct families.
• `bash -n spin-node.sh` clean.

Output of the new selection box (case 1):

```
╔══════════════════════════════════════════════════════════════╗
║ 🗳 Aggregator Selection ║
╠══════════════════════════════════════════════════════════════╣
║ subnet 0 → zeam_0, grandine_3 ║
║ subnet 1 → ethlambda_4 ║
║ subnet 2 → qlean_2 ║
╚══════════════════════════════════════════════════════════════╝
```