bcachefs: use SRCU-aware waits after dropping locks

[Komzpa]

Summary

• use trans_wait_event() for the btree write throttling wait after drop_locks_do()
• use trans_wait_event() for the journal-low-space wait before btree update reservation
• keep the normal bch2_trans_unlock() fast path from drop_locks_do(), while letting long waits drop SRCU through the existing transaction-aware wait helper

Context

This is a narrow follow-up to koverstreet/bcachefs#1069 and #659.

The earlier PR #659 handled the nocow-lock wait path. This patch covers the remaining direct closure_wait_event() calls inside drop_locks_do(trans, ...) scopes. drop_locks_do() already drops btree locks with bch2_trans_unlock(), but it does not drop the transaction SRCU read lock before the wrapped wait. Reusing trans_wait_event() keeps short waits cheap and escalates only if the wait outlives the transaction short-wait budget.

Validation

• git diff --check
• make generate_version && BINDGEN=/home/kom/.cargo/bin/bindgen LIBCLANG_PATH=/usr/lib/llvm-18/lib CARGO_BUILD_JOBS=4 make -j4 bcachefs
• safe staged kbuild module compile without unload/install side effects:
  make -C /lib/modules/$(uname -r)/build M=/home/kom/tmp/bcachefs-module-srcu-aware-wait-next modules -j4
• GitHub Actions: Nix Flake actions run 28428366238, 17/17 jobs passed on head bd0506446be2b8186975c1edce51db30a20274cf

[koverstreet]

Merged — cherry-picked onto master (e2097c2, authorship preserved, retitled btree:). Both waits sit inside drop_locks_do() with SRCU still held, so the indefinite closure_wait_event() stalled the grace period; trans_wait_event() is the SRCU-aware fix. Thanks!