ML predictions Phase 4: SARIMA spending forecast with dual confidence bands

Replaces unused Prophet dependency (unrunnable without cmdstan) with
SARIMA (statsmodels SARIMAX) as the primary spending forecast algorithm.
Strategy: SARIMA(1,1,1)(1,0,1,12) for 12+ months of data, ARIMA(1,1,1)
for 6-11 months, Holt-Winters for 3-5 months, simple average below that.

Adds 95% confidence bands (1.96σ) alongside existing 80% (1.28σ).
Extends forecast horizon from 3 to 6 months and actuals display from
6 to 12 months. Each category now carries an algorithm field surfaced
as a badge in the UI. Frontend chart shows both confidence tiers as
stacked bar overlays with a 3-month summary grid below.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/pyproject.toml

@@ -17,7 +17,6 @@ dependencies = [
     "qrcode[pil]>=8.0",
     "cryptography>=44.0",
     "yfinance>=0.2",
-    "prophet>=1.1",
     "statsmodels>=0.14",
     "numpy>=2.0",
     "scipy>=1.14",
@@ -52,6 +51,7 @@ build-backend = "hatchling.build"
 
 [tool.pytest.ini_options]
 asyncio_mode = "auto"
+asyncio_default_fixture_loop_scope = "session"
 testpaths = ["tests"]
 
 [tool.hatch.build.targets.wheel]