fix: memory-based reader paths drop records after blank lines (#122)

[stevehansen]

Summary

• MemorySliceLineSource.TryReadLine and MemoryReaderLineSource.TryReadLine returned false on blank middle-of-stream lines, terminating enumeration. Read / ReadAsSpan / ReadAsync don't have this bug because TextReader.ReadLine distinguishes blank \"\" from EOF null.
• TryReadLine now returns false only when position >= csv.Length at entry. Blank lines surface as empty MemoryText with return true, and the engine's existing SkipRow logic handles them (default predicate skips empties; user-overridden predicate may surface them).
• Adds three cross-path regression tests in Csv.Tests/EngineUnificationTests.cs.

Resolves #122.

Repro on master

```csharp
var csv = "a,b,c\r\n1,2,3\r\n\r\n4,5,6\r\n";
foreach (var row in CsvReader.ReadFromMemoryOptimized(csv.AsMemory()))
Console.WriteLine(row.Raw);
// Before: just "1,2,3" -- the "4,5,6" record is silently dropped.
// After: "1,2,3" then "4,5,6".
```

Pre-existing

Not a regression from #118 — this behavior was present in the pre-#118 `ReadLineOptimized` and `MemoryText.ReadLine(ref position)` extension. #118 preserved it verbatim per its consolidation scope. Gemini's review on #121 flagged it; filed as #122 because the fix requires its own correctness analysis around `SkipRow` semantics for blank lines.

Test plan

• `dotnet build` clean on netstandard2.0, net8.0, net9.0
• `dotnet test` — 172/172 passing (excluding the pre-existing flaky `Memory_AllocationComparison` GC-ratio test)
• `When_BlankLineInMiddleOfStream_Then_AllPathsContinueParsing` — all 5 paths return 2 records for `"a,b,c\n1,2,3\n\n4,5,6\n"` with default options
• `When_BlankLineAndSkipRowDisabled_Then_AllPathsReturnEmptyRecord` — same input with `SkipRow = (_, _) => false` surfaces the blank line as an empty record (3 records total: `1,2,3` + empty + `4,5,6`)
• `When_TrailingBlankLine_Then_AllPathsTerminateAfterLastRecord` — trailing blank line doesn't produce an extra record

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request updates the CSV parsing engine to handle blank lines more consistently. Key changes include modifying TryReadLine to return true for empty lines rather than terminating the read, and updating test utility methods to verify column existence before access. Three new test cases were added to validate the handling of blank lines in various scenarios, such as in the middle of a stream or at the end. I have no feedback to provide as there were no review comments to evaluate.