Collect and interpret Firecracker balloon statistics (/balloon/statistics) to detect guest memory pressure before reclaiming or adding memory

Documented steps

1. Install the balloon during VM setup (before InstanceStart) via PUT /balloon with amount_mib, deflate_on_oom and crucially a non-zero stats_polling_interval_s to enable statistics
2. curl --unix-socket /tmp/firecracker.socket -X PUT http://localhost/balloon -d '{"amount_mib":0,"deflate_on_oom":false,"stats_polling_interval_s":10}' - a non-zero interval enables virtio balloon stats updates every N seconds
3. Read the latest statistics snapshot: curl --unix-socket /tmp/firecracker.socket -X GET http://localhost/balloon/statistics
4. The payload includes target_pages/actual_pages (in 4K pages) and target_mib/actual_mib (device-held MiB) taken from device config space, plus VIRTIO_BALLOON_S_* metrics: SWAP_IN, SWAP_OUT, MAJFLT, MINFLT, MEMFREE, MEMTOT, AVAIL, CACHES, HTLB_PGALLOC, HTLB_PGFAIL
5. On guest kernels >= 6.12 additional pressure metrics appear: OOM_KILL, ALLOC_STALL, ASYNC_SCAN, DIRECT_SCAN, ASYNC_RECLAIM, DIRECT_RECLAIM; ASYNC_SCAN/ASYNC_RECLAIM rise from kswapd, DIRECT_SCAN/DIRECT_RECLAIM from processes stalled on allocation
6. Change the polling cadence later via PATCH /balloon with only stats_polling_interval_s