Summary

Change onDemandMonthly() in frontend/src/recommendations.ts to return the
provider-supplied r.on_demand_cost directly instead of reconstructing it from
monthly_cost + savings + (upfront_cost / months_in_term). When
on_demand_cost is missing or zero, render — instead of falling back to a
reconstructed value.

Why

PR #322 deliberately scoped the On-Demand Monthly column to a
reconstruction-only helper (its commit comment says: "It does NOT use
on_demand_cost even when available, because the column's purpose is to display
the reconstructed denominator so users can verify the formula against the raw
fields visible in the same row").

In practice, post-#277 / #312 / #324 every supported provider (AWS, Azure, GCP)
plumbs on_demand_cost through to the frontend, and effectiveSavingsPct()
already prefers the provider value when present (see
frontend/src/recommendations.ts:738-756 — hasOnDemand branch). The two
columns therefore disagree for any row where the formula reconstruction drifts
from the provider's billed on-demand price (rounding, partial-day proration,
provider-specific list price quirks, AWS On-Demand Capacity Reservations, etc.).

The product decision: surface the authoritative number from the provider,
not a teaching-tool reconstruction. Users who want to verify the formula can
still see monthly_cost + savings + upfront_cost/term in the row's other
columns and do the arithmetic themselves; making On-Demand Monthly match
effectiveSavingsPct's denominator is more important than making it
reconstructible.

Scope

• onDemandMonthly(): drop the reconstruction. Return r.on_demand_cost when
  it's a positive number; return null otherwise (so the cell shows —,
  consistent with the existing monthly_cost == null em-dash rendering).
• Update the doc comment on onDemandMonthly() to reflect the new contract.
• Update the unit tests in frontend/src/__tests__/recommendations.test.ts:
  • keep "prefer explicit on_demand_cost",
  • replace "fall back to reconstruction" tests with "return null when
    on_demand_cost is null/missing/zero",
  • the existing reconstruction-arithmetic tests can be deleted — they're
    asserting behaviour we're explicitly removing.

Out of scope

• effectiveSavingsPct keeps its existing fallback (it's the canonical
  Effective Savings % calculation and is needed for backward compatibility on
  legacy cached rows that pre-date fix(recommendations/aws): plumb on_demand_cost for canonical Effective Savings % (closes #303) #312/fix(recommendations/aws): flush stale on_demand_cost cache + parser visibility (closes #321) #324).
• No backend change.

Acceptance criteria

• onDemandMonthly() returns r.on_demand_cost when it's > 0.
• Returns null when r.on_demand_cost is null, undefined, or 0.
• Column cell renders — for null cases (same as monthly_cost-null today).
• Sort + numeric filter on on_demand_monthly still treat null as bottom
  (POSITIVE_INFINITY for sort, NaN for filter — already in place).
• Unit tests added/updated; full test suite green.
• No regressions in existing column ordering / sort / filter tests.

Related

• fix(recommendations): plumb provider on_demand_cost through to frontend (closes #274) #277 — plumbed provider on_demand_cost to frontend (AWS/Azure/GCP)
• fix(recommendations/aws): plumb on_demand_cost for canonical Effective Savings % (closes #303) #312 — used it in effectiveSavingsPct
• feat(frontend/recs): add 'On-Demand Monthly' column to recommendations table #322 — added the column with reconstruction-only logic (this issue overrides
  that design choice)
• fix(recommendations/aws): flush stale on_demand_cost cache + parser visibility (closes #321) #324 — flushed stale on_demand_cost JSONB cache