fix(streamable-http): wait for post_writer before yielding

[Epochex]

Fixes #1675.

What happened

streamable_http_client() started StreamableHTTPTransport.post_writer() with tg.start_soon(), then yielded (read_stream, write_stream) immediately.

Because the write side is a zero-buffer in-memory stream, there is a narrow startup window where the first requests right after initialize() can race with the writer task entering its stream contexts. That shows up as flaky/incorrect behavior on the first few calls (e.g. list_tools()).

What changed

• Start post_writer() with await tg.start(...) and have post_writer() signal readiness via TaskStatus.started(None).
• Add regression tests that open short-lived sessions and issue consecutive requests immediately after initialize().

Extra hardening

While iterating on the Windows lowest-direct matrix, one job surfaced an unclosed pipe transport warning during teardown. I added an explicit process.stdout.aclose() in stdio_client() teardown to make subprocess cleanup more robust.

Tests

• uv run --frozen pytest tests/shared/test_streamable_http.py -k 'no_race_on_consecutive_requests or rapid_request_sequence'
• uv run --frozen ruff check src/mcp/client/streamable_http.py tests/shared/test_streamable_http.py
• uv run --frozen pyright src/mcp/client/streamable_http.py
• uv run --frozen pytest tests/issues/test_1027_win_unreachable_cleanup.py tests/issues/test_129_resource_templates.py

[Epochex]

Hi @maxisbey — this fixes #1675 (StreamableHTTP startup race right after initialize). It waits for post_writer readiness before yielding streams and adds regression tests. I also tightened stdio teardown by closing stdout to avoid an unclosed pipe transport warning on Windows.

Local tests are listed in the PR body. Would appreciate a review when you have a moment.