Stop mis-routing Gloas submissions to the Electra-shaped validation RPC

[0w3n-d]

What this PR does

`SimulatorClient::sim_request_builder` picked the validation RPC method by fork:

```rust
let method = if fork == ForkName::Fulu { &self.sim_method_v5 } else { &self.sim_method_v4 };
```

Since #514 made `submit_block` able to decode Gloas-fork submissions (unrelated to this bug -- helix's own builder-submission wire shape didn't need to change for Gloas, so nothing stopped a Gloas submission reaching this point), a Gloas submission landing here would silently fall into the `else` branch and get validated via `sim_method_v4`, the Electra/Prague-shaped RPC. Wrong, and silent about it.

• `sim_request_builder` now returns `Option<(RequestBuilder, &str)>`, explicit about which forks it knows how to route (`Fulu` -> v5, `Bellatrix`/`Capella`/`Deneb`/`Electra` -> v4) and returning `None` for anything else (currently `Gloas`/`Heze`, plus the pre-merge forks that can't reach this path anyway).
• Added `BlockSimError::UnsupportedFork`, treated as temporary/non-demotable -- this is helix's own limitation, not the builder's fault.
• The one caller (`SimulatorTile::spawn_sim`) now reports that error and drops the submission instead of dispatching, mirroring the existing early-return pattern used for missing decoded data / hydration misses just above it.

What this PR deliberately does NOT do

Does not add real Gloas validation support to `helix-simulator` -- that needs a Gloas-aware EL client (reth has no Amsterdam/ePBS support in its tagged release yet, only on a devnet branch) and a new `validateBuilderSubmissionV6` path. Tracked as its own issue.

Tests

Written before implementation:

• `routes_fulu_to_v5`, `routes_electra_to_v4`: existing routing unchanged.
• `gloas_has_no_validation_method_yet`: the actual regression this PR closes.

Reviewer checklist

• CI (`lint`, `unit-test`) is green
• Nothing surprising