orchestrator: include current/max counts in ResourceExhausted error messages

[AdaAibaby]

Fixes #3583

Problem

When the orchestrator returns a ResourceExhausted gRPC error and the API logs "Node exhausted, trying another node", the attached error string carries no count information. All three ResourceExhausted sites were either missing the current value or missing both current and max:

// only max, no current:
"max number of running sandboxes on node reached (%d), please retry"

// no numbers at all:
"too many sandboxes starting on this node, please retry"
"too many sandboxes resuming on this node, please retry"

Changes

• packages/shared/pkg/utils/resizable_semaphore.go: Add Current() int64 and Limit() int64 accessors to AdjustableSemaphore
• packages/orchestrator/pkg/server/sandboxes.go: Include current=N, max=M in both running-sandboxes and starting-sandboxes error messages
• packages/orchestrator/pkg/server/utils.go: Include current=N, max=M in the resuming-sandboxes error message

After this change

"max number of running sandboxes on node reached: current=50, max=50, please retry"
"too many sandboxes starting on this node: current=10, max=10, please retry"
"too many sandboxes resuming on this node: current=10, max=10, please retry"

This makes the "Node exhausted, trying another node" log line actionable — operators can now tell at a glance whether the node is fully saturated or just throttled on in-flight starts.

Test plan

• Verify go build ./packages/shared/... passes
• Verify go vet ./packages/shared/pkg/utils/... passes
• Manually trigger a sandbox create against a node at its MaxSandboxesPerNode limit and confirm the log line now includes current/max counts