Reduce render argument allocations

[sebastienros]

The scope-management rewrite merged in #981 already removes the redundant render scope allocation. Named render arguments still allocate a temporary KeyValuePair[] on every execution, even though most calls pass only one or two arguments.

Approach

• Keep one- and two-argument values inline without temporary arrays.
• Rent cleared FluidValue[] storage only for three or more arguments and return it on both success and failure.
• Continue evaluating every expression and Assigned callback in the caller scope before publishing any argument into the isolated render scope.
• Add callback-order/error-cleanup coverage and a focused two-argument BenchmarkDotNet case.

Benchmark

BenchmarkDotNet 0.15.8, .NET 10.0.11 Arm64, Apple M4 Pro, in-process toolchain, 3 warmup and 5 measurement iterations:

Shape	origin/main	This change	Difference
Two named render arguments	207.2 ns / 744 B	212.1 ns / 688 B	-56 B (-7.5%)

The timing confidence intervals overlap (198.3-216.1 ns baseline and 207.8-216.5 ns after), so this is treated as allocation-neutral throughput rather than a speed improvement. A pool-only two-argument version measured 218.5 ns / 688 B; inline storage was retained for the common case, with pooling reserved for larger argument sets.

Validation

• Targeted interpreted and compiled render tests: 3 passed in each mode
• dotnet test --no-restore: 2,750 passed, 20 skipped, 0 failed
• dotnet test --no-restore /p:Compiled=true: 2,750 passed, 20 skipped, 0 failed
• Release build of Fluid: netstandard2.0, net8.0, net9.0, and net10.0 with no warnings or errors

This PR now targets main directly because #981 merged the preceding scope-management layer.

[sebastienros]

Closing because this performance work is investigatory only; commit caaeb4f remains available on the branch for traceability.