fix(cuda): sync memory subtree streams before drop (#2175)

In some edge case where right after we start `build_async` on the memory
merkle subtrees, if the program panics, then the order of drop could be
that we drop the `initial_memory` buffers on the default stream first,
while the `build_async` kernels are still running and using those
buffers. This leads to a deadloop. I fixed it by just forcing the drop
to drop subtrees first (which should sync their special streams) before
dropping `initial_memory`.

compare:
https://github.com/axiom-crypto/openvm-reth-benchmark/actions/runs/18733111153

docs(transpiler): clarify `itof` function definition (#2170)

feat(cuda): VPMM v3 and async app prover (#2165)

Co-authored-by: Jonathan Wang <[email protected]>
Co-authored-by: Ayush Shukla <[email protected]>

chore(cuda): BabyBear unified + CUDA_DEBUG + opener in natural order (#…

…2149)

Co-authored-by: Jonathan Wang <[email protected]>

chore: bump stark-backend to `v1.2.1-rc.2` (#2148)

includes openvm-org/stark-backend#131

chore(release): v1.4.0 (#2110)

chore: bump workspace to `v1.4.0-rc.10` (#2097)

chore: bump workspace to `1.4.0-rc.9` and stark-backend tag to `v1.2.…

…0-rc.7` (#2078)

@jonathanpwang We can use this PR for a future tag update

chore: bump workspace to `v1.4.0-rc.8` and add `"cuda"` to CLI (#2069)

Add `cuda` feature to CLI, which is passthrough of SDK's cuda feature

chore: bump workspace to `v1.4.0-rc.7` (#2063)