[Kiril1Shevtsov]

A compounded overnight coupon with a nonzero daily-compounded spread can forecast its future part using a bare discount-factor ratio, omitting the spread inside each daily factor. Keep every overnight period for this case and use the existing daily-product path. Zero spread and additive-after spread retain their current shortcuts.

Add a deterministic regression with expected products derived from separate business-day enumeration, explicit historical inputs and overnight deposit discount factors. It covers future, mixed and historical observations; positive and negative spreads; lockout; full and telescopic schedule requests; observation shift for full accrual; and unshifted partial intervals.

Validation against cf74c9f657150750dfbf00ed6fd4c0768bc5ce36:

• The new stock regression fails 120 assertions on the baseline and passes all 241 assertions with the fix.
• The final unfiltered stock suite passes 1,715 test cases and 619,978 assertions, with no failures or skipped cases.
• An independent future oracle passes all 251 checks, compared with 84 failures on the baseline. Another 48 boundary-control checks pass on both.
• git diff --check and tools/check_test_messages.py pass.

Native validation used Linux x86_64, GCC 14, Boost 1.87 and C++17, shared/default configuration, unity builds, -O1 -g0 -DNDEBUG and warnings as errors. The changed production and stock-test units were rebuilt and linked with hash-verified unchanged baseline objects.

The exact daily-product fallback has O(N) construction, storage and forecasting cost. The broader control matrix retains 16 separate compressed-lockout anchor failures, addressed independently in #3058. This PR does not include that fix or the separate today-fixing corrections. Daily-spread gearing remains one, as required by the existing contract.