[core] Audit and standardize error handling for Raylet IPC and Plasma store interactions

[codope]

Background

We just refactored PutInLocalPlasmaCallback to return Status and propagated failures through TaskManager instead of crashing. During shutdown, IOError from plasma puts is tolerated and logged; otherwise, we surface a system failure to the task owner. See #55367

We should apply this pattern consistently across other Raylet IPC and plasma callsites.

Goal:

Ensure all Raylet IPC and Plasma interactions:

• Return/propagate Status to callers instead of hard-crashing (no RAY_CHECK_OK on external IPC paths).
• Document expected Status codes per operation (e.g., OK/ObjectExists, ObjectStoreFull, IOError, Disconnected).
• Tolerate shutdown-specific IOErrors where appropriate (warn and continue), but treat non-shutdown failures as real errors handled at the task/owner level.
• Avoid status-message string matching; use status codes.

Scope (primary targets to audit):

• Core worker plasma paths in CoreWorker and CoreWorkerPlasmaStoreProvider:
  • PutInLocalPlasmaStore, Put, CreateExisting, SealExisting, Release, GetIfLocal, Contains, Delete/DeleteImpl, GetPlasmaUsage.
• Raylet IPC paths relevant to object lifecycle:
  • PinObjectIDs RPC callback handling, UpdateObjectLocation/pubsub paths.
• Higher-level consumers:
  • TaskManager return-object handling (done in current PR), ObjectRecoveryManager, FutureResolver, generator streaming HandleReportGeneratorItemReturns (evaluate best-effort vs strict propagation).
• Any remaining RAY_CHECK_OK immediately after Raylet/Plasma calls that can return expected runtime errors.