fix: reject internal missing days in crypto rotation backtests

[Pigbibi]

Scope

Reject an internal missing calendar day in the global daily panel before rotation simulation. This prevents multi-day price changes, signal lag and observation counts from silently collapsing into fewer daily steps.

Only two paths change: the existing live-pool simulator and its focused price tests. No dependencies, pins, workflows, providers or runtime adapters change.

Preserved boundaries

• No requirement for every symbol to have its entire historical price series; late-listed unselected symbols remain valid.
• No synthesized prices, cash rows or zero returns.
• Requested-window endpoints, existing CLI tail tolerance, signal lag, terminal valuation, fees and the existing all-cash empty-result behavior are unchanged.
• This closes only the internal-whole-day-gap subcase, not all of RA-10. Existing artifacts are not recomputed or replaced.

Verification

• Original base: six new gap cases failed with DID NOT RAISE.
• Local offline regression: 77 tests passed across five related test files; independently reproduced by the coordinator.
• Ruff 0.15.22 passed for both changed files in the coordinator existing test environment; git diff --check passed.
• Existing CryptoLivePool exporter -> synthetic CSV -> run_walk_forward with actual pinned QPK cd6edbb source: the consumer rejects the gap, persists zero successful result JSON files and writes no returns output. Independently repeated in a fresh temporary directory.
• Tests used OS network denial, an empty environment, UTC, temporary HOME and no bytecode/cache. No real market data, model, provider, account or broker operations were executed.

Natural PR CI is pending for this head. Engineering success is not research or runtime adoption evidence. No manual workflow dispatch or schedule change is requested.