SS-4 Add streaming DATA message store path

[vhafdal]

Summary

Adds an opt-in streaming message store path for SMTP DATA handling.

Why

The existing DATA flow reads the complete dot-block into a ReadOnlySequence<byte> before invoking message storage. That preserves compatibility but forces full-message buffering before custom storage can process content. This change lets storage implementations opt into streaming while leaving existing IMessageStore implementations untouched.

Changes

• Add IStreamingMessageStore, extending IMessageStore with a PipeReader-based save method.
• Update DataCommand to stream DATA content to stores that implement IStreamingMessageStore.
• Keep the existing buffered IMessageStore.SaveAsync(... ReadOnlySequence<byte> ...) path for all current consumers.
• Add a streaming dot-block pipe helper that removes dot-stuffing while preserving strict max-message-size enforcement.
• Add pipe-level tests and a server-level test proving streaming stores are selected.

Validation

• DOTNET_ROLL_FORWARD=Major dotnet test src/SmtpServer.Tests/SmtpServer.Tests.csproj
• Result: 120 passed, 1 skipped.

Notes

• This is an additive API; existing message stores do not need to change.
• BenchmarkDotNet benchmarks were not run; validation focused on behavior and compatibility tests.

Jira

• SS-4

[vhafdal]

Benchmark follow-up added in vhafdal:SmtpServer@5df5db2.\n\nCommand used:\nDOTNET_ROLL_FORWARD=Major dotnet run -c Release --project src/SmtpServer.Benchmarks/SmtpServer.Benchmarks.csproj -- --filter *DataStoreBenchmarks*\n\nShortRun result on Test3.eml (~2 MB):\n- BufferedMaterializing: mean 91.887 ms, allocated 2060.05 KB/op, Gen0/1/2 collections reported.\n- StreamingDrain: mean 90.000 ms, allocated 26.05 KB/op, no Gen0/1/2 collections reported.\n\nThe benchmark shows similar throughput for this local SMTP send path while avoiding roughly 2 MB of managed allocation per DATA message when the store consumes the stream directly.

[vhafdal]

Superseded by consolidated PR #284.