bug(dispatcher): FUEL_EXHAUSTED advises raising fuel_cap, a registry field the plugin documents as not implemented

## Symptom

Every `Edit` to a large `.factory/` artifact produces:

```
fail-closed: FUEL_EXHAUSTED: fuel cap of 20000000 units exhausted;
raise fuel_cap or reduce payload size
```

from `validate-factory-path-root`, `validate-input-hash`, and `validate-template-compliance`. Reproduced on every edit to two artifacts across a full working session (a 616 KB decision note and a ~515 KB story index). `validate-changelog-monotonicity` continues to run throughout.

## The defect is the remedy, not the exhaustion

The message instructs the operator to `raise fuel_cap`. **That field is not implemented.** From the shipped `hooks-registry.toml` (lines 681–687), verbatim:

```
# No fuel_cap field: the per-plugin fuel_cap registry field (ADR-035 §Decision 5,
# ADR-039 §Decision 2) is not yet implemented — ADR-039 Phase 1 (registry fuel_cap
# schema extension) has not shipped. ... Until then the only available lever is
# the global InvokeLimits::default() cap
```

So of the two remedies the error offers, one names an unimplemented field and the other — "reduce payload size" — is unavailable to an artifact that must grow as it accumulates content. Shrinking it requires writes, which trip the same hooks.

## The headroom has already been measured, and a safety claim retracted

`hooks-registry.toml` lines 674–677, verbatim:

```
# NOTE: max_bytes caps are
# NOT calibrated against the fuel budget — BC-5.39.010 v1.15 RETRACTED that claim
# as false (measured fuel consumption ~99.21% of the 10M-instruction budget at
# typical artifact sizes, i.e. near-exhaustion, not comfortable headroom).
```

**Deliberately not over-claiming this number:** that 99.21% was measured against a **10M** budget, while the live runtime cap is **20M**. I have not measured current consumption at typical sizes and do not claim the figure transfers. The two data points bracket the problem without extrapolation — near-exhaustion at typical sizes under 10M, and actual exhaustion at 616 KB under 20M.

## The existing mitigation is scoped to one filename while the budget is global

`validate-state-size.sh` (lines 123–128) blocks above 4000 lines and warns above 3500, with the reason `"WASM fuel exhaustion risk per D-442(e)"` — for `lessons.md` only.

Of the 38 shell hooks shipped in `hooks/`, exactly **2** reference a line count at all (`validate-state-size.sh`, `precompact-flush-prune.sh`). The 616 KB note that exhausts the budget is 4112 lines — past the threshold the plugin itself treats as dangerous — and sits in a directory with no guard.

## A hypothesis tested and disproved

Recorded so it isn't re-run: fuel use is **not** simply proportional to how much of the target file a hook reads. `validate-factory-path-root` is 59 lines with no whole-file read and exhausts; `validate-changelog-monotonicity` is 195 lines with a whole-file read and survives. All four route through `legacy-bash-adapter.wasm`, so the difference is per-script cost or early-exit structure, not the adapter's payload handling. I did not identify the actual cause; early exit is the untested candidate (`validate-changelog-monotonicity` likely returns immediately on a file with no changelog table).

## Asks

1. Ship ADR-039 Phase 1, or change the message so it stops naming an unimplemented field.
2. Emit a signal distinguishable from success when a hook is skipped for fuel. Right now a fuel-skipped validator and a clean pass are indistinguishable in the transcript, so validation coverage is silently lost on exactly the largest, most-edited artifacts.
3. Generalize the `validate-state-size` guard beyond `lessons.md` to any `.factory/` artifact past the same threshold, or document why `decisions/` and `stories/` are exempt.

## Environment

Plugin `1.0.0-rc.24`, `hooks-registry.toml` `schema_version = 2`, dispatcher `factory-dispatcher` (single native binary, all hooks per event), wasmtime 46.0.2. ADR-035/039/042 are cited by the registry but not shipped in the plugin, so I could not read them.

The error originates in the dispatcher's fuel exhaustion logic, likely in the WASM adapter or hook runner. Examine the code generating the FUEL_EXHAUSTED message (search for that string) and the registry handling for fuel_cap. Review the hooks-registry.toml schema and the ADR-039 references. Testing changes requires understanding the hook execution flow and the artifact validation pipeline.

Scritto dal modello di indicizzazione a partire dal testo della issue.