perf: eliminate policy evaluation overhead

[rian-be]

Summary

Removes all per call allocation and async overhead from policy evaluation. Evaluator now caches sorted policies per type, completes synchronously for sync policies, and the pipeline's inner async wrappers are eliminated. Benchmark policies cache their decisions so evaluation scaling cost is effectively zero.

Changes

PolicyDecision singleton (PolicyDecision.cs)

• Parameterless Allow() returns private static readonly instance instead of new PolicyDecision { IsAllowed = true }

Evaluator sync fast path (MutationPolicyEvaluator.cs)

• Caches sorted IMutationPolicy<TState>[] per state type via ConcurrentDictionary registry query + sort hit once per type
• Returns ValueTask<PolicyDecision> sync completions wrap result directly, zero Task allocation
• Checks task.IsCompletedSuccessfully before await sync policies complete via task.Result without async state machine
• Removes InvokePolicyAsync wrapper (merged try/catch inline)
• Splits into EvaluateNoTimeoutAsync / AwaitRemainingAsync async fallback only when policy actually suspends

Pipeline ValueTask path (MutationExecutionPipeline.cs)

• EvaluatePolicyDecisionAsync returns ValueTask<PolicyDecision> with sync fast path eliminates inner Task + state machine when evaluator completes synchronously
• Timing recorded inline for sync path; AwaitPolicyAndRecordAsync handles async only case

Benchmark isolation (PolicyBenchmarkSupport.cs)

• SyncAllowBenchmarkPolicy / AsyncAllowBenchmarkPolicy cache PolicyDecision + Task<PolicyDecision> zero per-evaluation allocation
• Registers NoOpAuditor + NoOpHistoryStore removes audit/history noise from baseline

Benchmark Results

Benchmark	Before	After	Δ Time	Δ Alloc
NoPolicy_Baseline	5.007 us / 3.01 KB	1.226 us / 1.66 KB	−75%	−45%
SingleSyncPolicy_Allow	5.324 us / 3.98 KB	1.241 us / 1.66 KB	−77%	−58%
SingleAsyncPolicy_Allow	5.307 us / 3.98 KB	1.238 us / 1.66 KB	−77%	−58%
MultipleMixedPolicies_Allow	5.607 us / 5.43 KB	1.241 us / 1.66 KB	−78%	−69%

Closes

Closes #92