fix(icicle): extend mutex scope to protect NTT domain operations

Fix race condition where the nttDomainMu mutex was released before
ReleaseDomain() and InitDomain() operations completed, allowing concurrent
threads to corrupt domain state.

Issue: When multiple ProvingKey instances with different domain sizes are
initialized concurrently, Thread A could begin InitDomain() while Thread B
simultaneously calls ReleaseDomain(), causing undefined behavior in the
ICICLE C++ backend.

Solution: Use defer to hold the mutex for the entire critical section,
including the async domain operations. This ensures atomic domain
initialization and prevents interleaved domain management.

Impact:
- Prevents silent domain corruption that could lead to invalid proofs
- Eliminates potential panics from concurrent domain operations
- Critical for production environments with concurrent circuit initialization

The mutex now protects:
1. Reading/updating nttDomainMaxByDevice map
2. ReleaseDomain() execution (if needed)
3. InitDomain() execution
4. Waiting for domain operations to complete

Author: p4u

backend/accelerated/icicle/groth16/bls12-377/icicle.go

@@ -110,12 +110,13 @@ func (pk *ProvingKey) setupDevicePointers(device *icicle_runtime.Device) error {
 	needDomainRelease := false
 
 	nttDomainMu.Lock()
+	defer nttDomainMu.Unlock()
+	
 	currentMax := nttDomainMaxByDevice[device.Id]
 	if requestedDomainSize > currentMax {
 		needDomainRelease = currentMax != 0
 		nttDomainMaxByDevice[device.Id] = requestedDomainSize
 	}
-	nttDomainMu.Unlock()
 
 	if needDomainRelease {
 		releaseDomain := make(chan struct{})