lib/nvmf: remove vfio-user migration code

The live migration support in vfio-user has not been in a working state
for some time. The support in libvfio-user has been replaced by a
v2-style interface, which is significantly different. The qemu client
merged in QEMU 10.1 has no live migration support currently, but when it
does it will be v2 based.

Remove all the current live migration support. This also lets us refresh
libvfio-user, as we no longer need to stay on the special "spdk" branch.

There is another reason for this: the (legacy) migration region was
mappable by QEMU, which made all region writes non-posted. This client
behaviour is to account for read-after-write races with mappable versus
non-mappable regions, but a) the migration region isn't mappable by the
guest VM anyway, and b) this is theoretical for NVMe devices and
doorbell writes anyway. So removing the migration region altogether as
we do here is a simple way to get (faster) posted doorbell writes.

Submodule libvfio-user 5b82b9601...f922ada35:
  > docs: use custom image for gpio test (#808)
  > spruce up shadow ioeventfd documentation (#810)
  > Update SPDK docs (#811)
  > Update runs-on past ubuntu-20.04 (#813)
  > Add further sanity checking of hdr->error_no (#805)
  > samples: keep coverity quiet (#804)
  < lib/libvfio-user: Fix calloc-transposed-args reported by gcc >= 14 (#802)
  < fix VFIO_USER_DEVICE_GET_REGION_IO_FDS allocation (#785)
  < correct test_device_set_irqs.py
  > lib/libvfio-user: Fix calloc-transposed-args reported by gcc >= 14 (#802)
  > workaround centos:7 issue (#803)
  < bump MAX_DMA_REGIONS (#800)
  > bump MAX_DMA_REGIONS (#800)
  > docs/qemu: add missing -numa argument (#799)
  > improve documentation on testing with qemu (#797)
  > docs: correct markdown links (#796)
  > remove arch OS github action (#798)
  > Remove dangling quotes from vfio-user.rst (#795)
  > docs: document SPDK dependency (#794)
  < correct IRQ range check (#791)
  > Support 64 bits and prefetchable BARs (#792)
  > correct IRQ range check (#791)
  > Fix DMA message size calculation (#788)
  > make arch CI non-required (#789)
  > replace bcopy() with memcpy() (#786)
  > fix VFIO_USER_DEVICE_GET_REGION_IO_FDS allocation (#785)
  > test_negotiate: add migration testing (#781)
  > fix: minor memory bugs #784
  > Add some notes on live migration version and SPDK usage. (#783)
  > adapt to VFIO live migration v2 (#782)
  > Pass server->client command over a separate socket pair (#762)
  > Describe the twin-socket feature in the spec (#775)
  > update SDPK version spdk.md (#769)
  > Introduce client object in python tests (#772)
  > Prepare python test helpers for receiving commands (#774)
  > Construct server capabilities using json-c (#771)
  > Replace protocol header flags bit field with mask (#773)
  > fix: incorrect number of dirty pages printed (#766)
  > Make debian-11, debian-12 and arch-202307 required pull request jobs
  > Revise iovec_t.__eq__ and vfu_dma_info_t.__eq__ to fix flake8 E721
  > Add debian 11, 12 and arch linux jobs to pull_request workflow
  > Bump Github Actions version to v3
  > Introduce close_safely helper function (#763)
  > Allow adding MSI capability via vfu_pci_add_capability (#758)
  > Add another lspci test output variant (#761)
  > Fix MSI-X capability write logging opposite status (#759)
  > fix: server sample not marking dirty pages (#748)
  > docs: document test debugging with GDB (#756)
  > docs: document DMA message dirty handling (#755)
  > fix: CRC calculation in client sample (#750)
  > refactor: remove private includes from server.c (#752)
  > Fix address calculation for message-based DMA (#740)
  > fix build instructions for samples (#739)
  > Fix some python tests (#737)
  > fix err/req irq fd issue (#731)

Signed-off-by: John Levon <john.levon@nutanix.com>
Change-Id: I725c90ee7f15943a03c556f7f392c299f75ce2c6
Reviewed-on: https://review.spdk.io/c/spdk/spdk/+/26595


Reviewed-by: Changpeng Liu <changpeliu@tencent.com>
Tested-by: SPDK Automated Test System <spdkbot@gmail.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <jim.harris@nvidia.com>