refactor: persist EWMA only on LLM failure, dedup EWMA helper

Audit items #7 and #5.

#7 — BrentPricePredictor::generatePrediction previously wrote both an
EWMA row and an LLM row to price_predictions on every run. The
downstream OilSignal already prefers llm_with_context > llm > ewma, so
the EWMA row was dead weight 95% of the time. Now we try LLM first; if
it returns null (no API key, parse failure, etc.) we compute and persist
EWMA as a real fallback. This also avoids redundant work on the success
path.

Updated the "stores both" test to "stores only LLM" — asserts no EWMA
row is written when the provider succeeds.

#5 — BrentPricePredictor and AnthropicPredictionProvider both had
byte-identical computeEwma() methods with identical EWMA_ALPHA = 0.3
constants. Extracted to App\Services\Ewma::compute() and dropped both
private methods + their alpha constants.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tests/Unit/Services/BrentPricePredictorTest.php

@@ -61,7 +61,7 @@ it('returns null when fewer than 14 prices are available for EWMA', function ():
     expect($this->predictor->generateEwmaPrediction($prices))->toBeNull();
 });
 
-it('stores both EWMA and LLM predictions when provider succeeds', function (): void {
+it('stores only the LLM prediction when the provider succeeds', function (): void {
     seedPrices(20);
 
     $this->provider->shouldReceive('predict')->once()->andReturn(new PricePrediction([
@@ -76,7 +76,8 @@ it('stores both EWMA and LLM predictions when provider succeeds', function (): v
     $prediction = $this->predictor->generatePrediction();
 
     expect($prediction->source)->toBe(PredictionSource::LlmWithContext)
-        ->and(PricePrediction::count())->toBe(2);
+        ->and(PricePrediction::count())->toBe(1)
+        ->and(PricePrediction::where('source', PredictionSource::Ewma)->count())->toBe(0);
 });
 
 it('falls back to EWMA when provider returns null', function (): void {