feat(profiler): follow forked process trees

[btlqql]

Summary

Resolves #335 by adding opt-in automatic process-tree tracking to the native profiler.

• track forked and cloned descendants through sched_process_fork, sched_process_exec, and sched_process_exit
• inherit the root profiler's tracking eligibility for descendants
• keep surviving descendants traceable after parent/root exit, including non-leader thread exec PID migration
• add configurable process limits, global fork-rate protection, dynamic BPF map sizing, cleanup, and coverage telemetry
• wire the feature into native CPU and all native memory modes
• add CLI validation, bilingual documentation, and a deterministic lifecycle reference model with concurrency tests

Issue acceptance coverage

• Fork/clone tracking: descendants are added automatically from scheduler lifecycle events.
• Context inheritance: tracked descendants use the same native profiler configuration and filters as the root target.
• Batch-creation stability: bounded maps, reservation-based capacity enforcement, and fork-rate protection prevent unbounded growth during fork storms.
• Lifecycle cleanup: exit events remove process state; exec migration handles Linux PID changes without leaving stale keys.
• Observability: final coverage statistics report accepted, rejected, active, exited, limited, and migration counts.

Compatibility and safety

• preserves Linux 4.18 and BPF -mcpu=v1 compatibility by using discarded-result legacy XADD operations
• enforces the configured process cap through reservation plus the kernel map capacity
• uses non-preallocated PID storage and shrinks the disabled map to one entry
• detaches programs before reading final coverage statistics

Validation

• go test -count=10 ./internal/profiler/forktrack
• go vet ./internal/profiler/forktrack
• Linux cross-compilation of tests for ./internal/bpf, ./cmd/profiler/provider, and ./cmd/profiler
• git diff --check

Runtime eBPF attachment was not executed locally because the development host is Windows; Linux compilation and static BPF compatibility checks pass.