fix(binance): expose safe startup diagnostics

[Pigbibi]

Problem

Binance live startup failures were reduced to runtime_setup_failed stage=strategy_load error_type=RuntimeError, so the operator could not distinguish a recovery-control read failure from a later validation or authorization rejection.

Change

• Classify only the recovery-control read boundary as the fixed safe reason recovery_control_read_failed.
• Reuse the startup validator's allowlist for safe exception reason codes, including RuntimeError subclasses carrying a fixed reason_code.
• Keep validation and authorization failures on their existing reason codes; unknown exceptions remain generalized.
• Show the machine reason code in runtime logs and a fixed human-readable Chinese or English explanation in notifications.
• Add offline stub coverage and document the diagnostic boundary.

This improves diagnosis only. It does not claim to identify the historical first failure, activate recovery, or restore a real trading cycle. There is no retry, recovery-state change, broker action, or order path.

Validation

• tests/test_reconciliation_recovery.py: 25 passed
• tests/test_validate_runtime_startup.py: 24 passed
• tests/test_main_runtime_error_notification.py: 12 passed, 2 subtests passed
• Total: 61 passed, 2 subtests passed
• Ruff and diff checks passed in the pinned validation environment.
• validate_only run 34998046567 on SHA e66ec910ec85536f9e7bfdafeaaa7593cb580f61 passed startup validation with execution_permitted=false; it did not execute a full cycle.

No deployment, live dispatch, merge, or production/runtime change is included.