fix(llm): cap provider response bodies

[CryptoJones]

Summary

• Send LLM requests with HttpCompletionOption.ResponseHeadersRead.
• Reject declared oversized LLM success responses before reading the body.
• Stream success and error response bodies with hard byte caps.
• Preserve concise logging for oversized upstream error bodies without allocating the full payload.
• Add isolated tests for normal completions, declared oversized success responses, streamed oversized success responses, and oversized error bodies.
• Bump version to 1.7.9.

Closes #48

Verification

• dotnet test api/ApplyTrack.slnx --filter "FullyQualifiedName~OpenAiCompatibleLlmClientTests|FullyQualifiedName~LlmConfigTests" --no-restore
• dotnet build api/ApplyTrack.slnx --no-restore
• dotnet format api/ApplyTrack.slnx --verify-no-changes --no-restore --include api/ApplyTrack.Api/Llm/OpenAiCompatibleLlmClient.cs api/ApplyTrack.Api.Tests/OpenAiCompatibleLlmClientTests.cs
• uv run --extra dev pytest -q
• git diff --check

Full .NET integration tests require the Podman/Testcontainers socket and should run in CI or via ./scripts/test-dotnet-podman.sh on a Podman-enabled host.