0.4.0

• Refactored code to use stack memory as much as possible, and avoid allocating/freeing memory. Even under intense load (200+ MBit/s), memory usage stays below 10MB.
• When possible, the code uses poll_fn to process multiple I/O operations in parallel - while still using a single-threaded version of Tokio.
• Improved bandwidth and throughput for the IKEv2 flow.
• Fixed an issue where sometimes shutdown never completed.
• Simplified the VPN client state machine.
• Bumped up dependency versions.
• Other small improvements.