perf(mcp): eliminate SSE buffer churn in streaming parser

[G7CNF]

Summary

• replace per-chunk trailing-slice reallocation (buffer = buffer[remaining_start..].to_string()) with in-place compaction (buffer.drain(..remaining_start))
• keep SSE parser behavior unchanged while removing repeated allocations in the hot streaming loop

Why

Issue #1142 identified avoidable allocation churn in parse_sse_response for large/many chunks. Reallocating a new String each loop iteration is unnecessary and can degrade throughput.

This change keeps the same parser semantics but reuses the existing buffer capacity.

Closes #1142

Validation

• cargo fmt --all
• cargo test -p ironclaw tools::mcp::http_transport::tests::test_http_transport_creation -- --nocapture
• cargo clippy -p ironclaw --all-targets -- -D warnings

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[henrypark133]

Review: good small perf cleanup

This keeps the SSE parser behavior the same while removing avoidable allocation churn in the hot loop.

Positives:

• drain(..remaining_start) is a straightforward replacement for the previous trailing-slice reallocation.
• The change is extremely small and easy to reason about.
• Meaningful CI is green.

No blocking issues from my side.