Track nested-lambda closure captures during flat-expression construction

[Copilot]

FlatExpression was indexing lambda/block/control-flow nodes, but it was not carrying forward the closure-shaping information that TryCollectInfo derives for nested lambdas. This change makes flat-expression construction record captured outer parameters/variables directly, so nested-lambda closure metadata is available without a separate collect pass.

• Add closure-capture metadata to ExprTree

  • Introduce LambdaClosureParameterUsage and ExprTree.LambdaClosureParameterUsages
  • Record, per lambda, which outer parameter/variable identities are used from outside the lambda’s own parameter list

• Collect captures during flat construction

  • Run capture analysis when building lambdas via the direct ExprTree.Lambda(...) API
  • Run the same analysis when flattening existing expression trees via ExprTree.FromExpression(...)
  • Handle the same scoping shapes that matter for closure creation:
    • nested lambdas
    • block-local variables
    • catch variables / filters
    • propagated captures through multiple nesting levels

• Preserve identity-based mapping

  • Store the shared parameter/variable identity (ExprNode.ChildIdx) so closure analysis stays aligned with the existing flat-tree identity model
  • Deduplicate repeated uses of the same captured parameter within a lambda

• Add focused coverage

  • Direct flat construction: captured outer parameter
  • Multi-level nesting: propagated capture into intermediate lambdas
  • Captured block variable
  • ToFlatExpression() conversion from System.Linq.Expressions

Example:

var fe = default(ExprTree);
var x = fe.ParameterOf<int>("x");

var inner = fe.Lambda<Func<int>>(x);
fe.RootIndex = fe.Lambda<Func<int, Func<int>>>(inner, x);

// inner lambda captures outer x
var usage = fe.LambdaClosureParameterUsages[0];
Console.WriteLine((usage.LambdaIndex, usage.ParameterId));

[dadhi]

The possible range of indexes is 2 bytes. Please use it here for indexes as well. Use explicit zyruct kayout to avoid padding.
Add xml doc comments for the public fields. The ci is failing.
Use the consistent naming for indexes - if we are using Idx in FE nodes, use Idx here indtead of index.

[Copilot]

Addressed in 7e61241 with the compact ushort/Idx update, and followed up in 772ea38 with the XML docs and explicit note about the shared 16-bit FE index range.

[dadhi]

Use shorts or better ushort for all stacks/lists with indexes here and below. Why are we wating 2 bytes per index?

[Copilot]

Addressed in 7e61241 by switching the new capture-analysis index stacks/lists to ushort, with the follow-up range/documentation cleanup in 772ea38.

[dadhi]

Ushort instead of int for each index operating list.

[Copilot]

Addressed in 7e61241 by using ushort for the new index-tracking lists and compact capture records, with the final range/documentation touch-up in 772ea38.

[dadhi]

@copilot please fix this warning from ci:

Warning: D:\a\FastExpressionCompiler\FastExpressionCompiler\test\FastExpressionCompiler.IssueTests\Issue500_IndexOutOfRangeException_with_value_objects_implicit_conversions.cs(27,85): warning CS8632: The annotation for nullable reference types should only be used in code within a '#nullable' annotations context. [D:\a\FastExpressionCompiler\FastExpressionCompiler\test\FastExpressionCompiler.IssueTests\FastExpressionCompiler.IssueTests.csproj::TargetFramework=net8.0]

[Copilot]

@copilot please fix this warning from ci:

Warning: D:\a\FastExpressionCompiler\FastExpressionCompiler\test\FastExpressionCompiler.IssueTests\Issue500_IndexOutOfRangeException_with_value_objects_imp...

Fixed in c388121 by enabling nullable context in the shared Issue500 test file so the ? annotations no longer produce CS8632 across the issue-test target frameworks.