Emit MCP progress notifications

[Widthdom]

Summary

• Emit MCP notifications/progress for index and backfill_fold when _meta.progressToken is supplied.
• Route stdio progress as out-of-band frames and HTTP progress through the existing /events SSE stream so POST responses still return the final tool result.
• Document the progressToken behavior in MCP tool descriptions and add a bilingual changelog fragment.

Validation

• dotnet test tests/CodeIndex.Tests/CodeIndex.Tests.csproj --filter "FullyQualifiedName~McpServerTests.RunAsync_IndexWithProgressToken_EmitsProgressNotificationBeforeResult|FullyQualifiedName~McpServerTests.RunAsync_NonStreamingIndexWithProgressToken_ReturnsFinalResultWithoutProgress|FullyQualifiedName~HttpMcpTransportTests.HttpTransport_IndexWithProgressToken_EmitsProgressOnEventsStreamAndReturnsResult"
• dotnet test tests/CodeIndex.Tests/CodeIndex.Tests.csproj --filter "FullyQualifiedName~HttpMcpTransportTests"
• dotnet test tests/CodeIndex.Tests/CodeIndex.Tests.csproj --filter "FullyQualifiedName~McpServerTests"
• dotnet run --project tools/CodeIndex.Changelog -- check
• dotnet test was also run; net9.0 passed, and net8.0 had one transient trimmed-publish file lock in QueryCommandRunnerTests.RunPublishedTrimmedCli_SearchSupportsCSharpRazorAliases(lang: "cshtml"). Re-running that net8.0 test passed.

Documentation / Changelog

• Updated MCP tool descriptions for index and backfill_fold.
• Added changelog.d/unreleased/1684.fixed.md.

Review

• Codex adversarial review found an HTTP one-shot response regression in the first implementation.
• Addressed by sending HTTP progress over /events SSE instead of the POST response writer.
• A second Codex review attempt was blocked by the local Codex usage limit.

Fixes #1684