test(research): characterize timing and cost limitations

[Pigbibi]

Scope

• Add 9 deterministic characterization tests for daily lookahead, same-row signal/fill, stale forward-filled quotes, trend and BTC DCA cash-buffer accounting, cost-free buys/sells, and missing pre-fill equity anchor.
• Document current semantics vs future acceptance, uncomputed timing/cost validation, and explicit research comparability limits.
• ONLY tests and documentation. The backtest model and all production/runtime/strategy/notification code are unchanged.

Relates to #366. Does NOT close #366: fee/slippage accounting and causal timing remain unresolved. Passing tests characterize existing defects; they do not claim a corrected model or real financial returns. First-fill-loss acceptance is documented as unimplemented because today's zero-cost model cannot generate that fixture naturally.

Verification

• Exact published head: ec485c2504468a160cf19794cddd7ff1f8538556; independently verified with git ls-remote.
• CI run 38135975559: terminal SUCCESS, including compile, repository-wide ruff, QPK pin/lock consistency and full pytest suite.
• 9 focused local unittest tests and repository-wide ruff: passed.
• Full local strictly offline pytest: 1504 passed, 208 subtests passed; one existing uv-lifecycle dependency-install fixture failed because its build dependencies were not all cached. Full local unittest: 486 run, same one dependency-cache error. This environment limitation was not hidden by skipping the fixture or modifying source; CI passed it.
• Independent read-only review: no actionable findings.

No merge or deployment is requested. This remains a draft.