Add per-process filesystem mount namespaces and working directories #191

Summary

Extend the filesystem mount model so processes can have an independent view of mounted filesystems and a well-defined current working directory.

Scope

• Define mount-namespace ownership, creation, inheritance, and teardown.
• Specify whether namespace changes are shared or private across fork/clone operations.
• Track a process working directory using a stable VFS reference rather than a raw path string.
• Define path resolution relative to the process root and current working directory.
• Prevent unmounting resources that are still referenced by active paths or file handles.

Acceptance criteria

• Tests cover namespace isolation, inheritance, path resolution, and teardown.
• Relative and absolute paths resolve consistently across process boundaries.
• Reference lifetimes prevent use-after-free during unmount.
• Document the initial namespace-sharing policy and unsupported mount operations.

Engineering constraints

• Keep the kernel Rust-first, no_std, and x86_64-compatible.
• Integrate with existing Rustalis subsystem interfaces; do not create a parallel implementation.
• Document ownership, initialization order, failure behavior, and any architecture-specific unsafe invariants.
• Add host-side tests for hardware-independent logic and QEMU coverage for hardware paths where practical.
• Do not claim verification that has not actually been run.