fix(volumes): broadcast volume directory creation to all cluster nodes

[AdaAibaby]

Problem

When a volume is created via POST /volumes, the directory is only
created on a single randomly-selected orchestrator node
(executeOnOrchestratorByClusterID uses rand.Shuffle and returns on
the first success). Sandbox placement (PlaceSandbox) is independent
and selects nodes based on CPU/RAM availability, with no awareness of
which node holds the volume directory.

This causes an intermittent failure: sandboxes that land on the same
node as the volume get a working NFS mount; sandboxes that land on any
other node fail to mount because the NFS proxy calls
syscall.Mount(path, path, MS_BIND) on a path that does not exist on
that node.

A second, related bug exists in envd/internal/api/init.go: when two
concurrent /init requests race on isMountingNFS, the second request
silently returns nil, causing the sandbox to start successfully with
no NFS mount.

Changes

volume_util.go

• Add executeOnAllClusterNodes: fans out a gRPC call concurrently to
  every ready node in the cluster using errgroup, returning a combined
  error if any node fails.

volume_create.go

• createVolume now calls executeOnAllClusterNodes instead of
  executeOnOrchestratorByClusterID, so the volume directory is
  created on every orchestrator node at creation time.

volume_delete.go

• deleteVolume likewise broadcasts to all nodes to clean up the
  directory everywhere it was created.

init.go

• setupNFS: when isMountingNFS CAS fails (concurrent request),
  return an explicit error instead of nil so the caller receives a
  400 and can retry, rather than silently succeeding with no mount.

Testing

• go build ./packages/api/... ./packages/envd/... packages. ✅
• go vet ./packages/api/internal/handlers/... internal. ✅
• go test internal. ✅
• Manual: on a staging cluster with ≥2 orchestrator nodes, creating a
  volume and then creating 10 sandboxes with that volume mount all
  succeed with a valid NFS mount (mount | grep nfs non-empty in every
  sandbox).

[cla-bot]

We require contributors to sign our Contributor License Agreement, and we don't have @adababys on file. You can sign our CLA at https://e2b.dev/docs/cla . Once you've signed, post a comment here that says '@cla-bot check'

[qodo-code-review]

ⓘ You've reached your Qodo monthly free-tier limit. Reviews pause until next month — upgrade your plan to continue now, or link your paid account if you already have one.

[AdaAibaby]

@cla-bot check

@cla-bot check

[cla-bot]

We require contributors to sign our Contributor License Agreement, and we don't have @adababys on file. You can sign our CLA at https://e2b.dev/docs/cla . Once you've signed, post a comment here that says '@cla-bot check'

[cla-bot]

The cla-bot has been summoned, and re-checked this pull request!

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bce5e5094d

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Do not cancel remaining node RPCs on first failure

Using errgroup.WithContext here cancels wgCtx as soon as one node returns an error, so other in-flight or not-yet-started node RPCs can be aborted before they execute. That breaks the intended “run on all nodes” behavior and can leave partial state (for example, deleteVolume may skip cleanup on healthy nodes after one early failure). Run goroutines under a non-canceling group and aggregate errors after all node attempts complete.

Useful? React with 👍 / 👎.

[gemini-code-assist]

Code Review

The executeOnAllClusterNodes function incorrectly uses errgroup.WithContext, which cancels all operations on the first error, preventing full cleanup during deletions, and returns success when no nodes are ready.

[cla-bot]

We require contributors to sign our Contributor License Agreement, and we don't have @adababys on file. You can sign our CLA at https://e2b.dev/docs/cla . Once you've signed, post a comment here that says '@cla-bot check'

[adababys]

We require contributors to sign our Contributor License Agreement, and we don't have @adababys on file. You can sign our CLA at https://e2b.dev/docs/cla . Once you've signed, post a comment here that says '@cla-bot check'

@cla-bot check

[cla-bot]

The cla-bot has been summoned, and re-checked this pull request!