[PE] Calculated field state is not initialized - Script CF on Device Profile

Describe the bug

A Script-type Calculated Field defined on a Device Profile intermittently fails during normal runtime operation with:

Calculated field state is not initialized!

java.lang.NullPointerException: Cannot invoke "org.thingsboard.server.service.cf.ctx.state.CalculatedFieldScriptEngine.executeScriptAsync(Object[])" because "expression" is null

The error is not limited to newly created or newly assigned devices — it occurs during ongoing operation for multiple devices that already share the same device profile and were previously processing telemetry without issues.

Steps to Reproduce

1. Create a Script-type Calculated Field on a Device Profile (in our case "V-Matic"), with:
   • count — Time series rolling, current entity, key adminProdProcess...
   • designSpeed — Attribute, current entity, key adminMachDesignS...
   • state — Time series rolling, current entity, key statusStateCurrent
2. Let telemetry flow normally to multiple devices assigned to this profile. The CF works correctly for a period of time.
3. At some point (not tied to a specific reproducible trigger we've identified yet), the CF starts throwing the NPE above for telemetry from several devices under the same profile simultaneously.
4. Workaround: Opening the Calculated Field in the UI, making a trivial change (or no change) and saving it again immediately resolves the issue — the CF resumes working correctly for a period of time before the error reappears.

Expected behavior

The Calculated Field context (CalculatedFieldCtx, including the compiled TBEL expression/engine) should remain valid for the lifetime of the CF configuration and should not need to be manually re-saved to keep functioning.

Observations / Suspected root cause

• Since the CF is defined at Device Profile level, all devices under that profile share the same CalculatedFieldCtx (per the CalculatedFieldManagerMessageProcessor architecture, which is per-tenant and maintains one context per CF).
• The fact that re-saving fixes it for all affected devices at once, and it later breaks again for multiple devices at once, strongly suggests the shared context itself is being invalidated/evicted or loses its compiled expression/script engine at some point during runtime — rather than an issue with any specific device's local state.
• This looks distinct from the NPE hardening shipped in PE v4.3.1.3 ("Hardened calculated field owner init against NPE and reduced DB load"), which appears to target owner (tenant/customer) resolution during first-contact initialization, not runtime context invalidation for profile-level Script CFs.

Environment

• ThingsBoard Version: 4.3.1.3 PE
• Calculated Field type: Script (TBEL)
• Scope: Device Profile-level CF (not single-device)
• Arguments: current-entity Time series rolling (x2) + current-entity Attribute (x1)
• Cluster setup: (please fill in: monolith vs. microservices/Kafka queue mode, number of rule-engine instances)

Additional context

• Error recurs across multiple devices under the same profile at overlapping times, not just isolated/new devices — pointing to a shared-context problem rather than per-device initialization race.
• (please attach: exact timestamps of failures vs. any CF edits, rule-engine restarts, deployments, or cluster rebalancing events if available; full stack trace; server logs around the failure window)