fix(whatsrust): avoid lifecycle handler registration deadlock

[Andiveli]

Problem

Startup can deadlock when C_Connect registers event handlers while holding the lifecycle mutex and the production registration callback calls clientSnapshot, which re-enters that mutex.

Proposed solution

Coordinate once-only registration under the lifecycle mutex, execute the registration callback outside it, and allow a panic to reset registration so a later attempt can retry. Add bounded behavior and concurrency regression coverage.

Acceptance criteria

• Registration callbacks never execute while the lifecycle mutex is held.
• Concurrent callers register handlers exactly once.
• A panicking registration does not permanently mark handlers as registered.
• Focused Go tests, go test ./..., go vet ./..., formatting, and whitespace checks pass.