perf: drop the pool dance from MemorySliceLineSource.Concat (#119)

[stevehansen]

Summary

• The pre-RFC: Unify the four CsvReader read loops behind one JIT-devirtualized engine #118 ConcatenateMemory (and its post-RFC: Unify the four CsvReader read loops behind one JIT-devirtualized engine #118 successor MemorySliceLineSource.Concat) rented a buffer from CharArrayPool, copied the three input segments in, then immediately allocated a fresh char[] of exact size and copied again — defeating the pool entirely and paying two allocations + three copies where one allocation suffices.
• Replace with string.Concat(head.Span, newLine.AsSpan(), tail.Span). Single contiguous allocation, no pool churn, no second copy. Materialized string is captured into out string? combined so it's not discarded.
• Drop the now-unused CsvMemoryOptions field and constructor parameter on MemorySliceLineSource; CsvReader.ReadFromMemoryOptimizedImpl updated accordingly.

Resolves #119.

Allocation profile

Path	Before	After
Per multiline concat	1 pool rent + 1 new char[totalLength] + 3 CopyTo	1 string.Concat (one alloc, internal contiguous copy)

Net: one fewer allocation, two fewer copies, no ArrayPool overhead, on every multiline continuation step of ReadFromMemoryOptimized.

Notes

• This was deliberately preserved verbatim by RFC: Unify the four CsvReader read loops behind one JIT-devirtualized engine #118 (approved deviation .NET Core support #4) to keep that PR scoped to consolidation. Fixed now per the follow-up.
• MemoryReaderLineSource.Concat already used StringHelpers.Concat (which internally calls string.Concat); only MemorySliceLineSource had the anti-pattern.
• OptimizedRowFactory.Create still ignores rawString, so the captured combined string isn't consumed downstream — but the cost is zero now that we've materialized it anyway.

Test plan

• dotnet build clean on netstandard2.0, net8.0, net9.0
• dotnet test — 169/169 passing (excluding the pre-existing flaky Memory_AllocationComparison GC-ratio test)
• Existing multiline tests in Csv.Tests/EngineUnificationTests.cs cover the ReadFromMemoryOptimized multiline path; they continue to pass without modification

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request simplifies the Concat method within the MemorySliceLineSource struct by replacing manual buffer pooling and copying logic with a call to string.Concat. As a result, the CsvMemoryOptions dependency has been removed from the struct and its constructor. I have no feedback to provide as there were no review comments.