Add an opt-in HTTP/2 connection soft limit

[thliu21]

HTTP/2 requests currently queue behind the server's MAX_CONCURRENT_STREAMS limit even when the server could handle more connections. This adds ConnectionPool.concurrentHTTP2ConnectionsPerHostSoftLimit for #567, defaulting to 1. With a server limit of one stream and a connection limit of two, two requests can run concurrently while a third remains queued.

This is a draft for API and behavior alignment.

Proposed behavior

• Reuse available streams, then expand on queued demand, waiting for a relevant pending connection attempt before expanding further.
• Count active, starting, and backing-off connections toward the soft limit; exclude draining connections.
• Allow a required event loop to exceed the limit to establish its first usable connection. Preserve the default's existing protocol-migration exception of one connection per event loop.
• Reevaluate queued demand after SETTINGS, retries, GOAWAY, protocol migration, and connection closure, including capacity released on another event loop.

The setting is also available through Swift Configuration. This proposal expands only after server-advertised stream capacity is exhausted; distributing requests across more connections before saturation is outside this patch.

Validation

On macOS with Swift 6.3.2:

• swift test -j 4 --filter 'HTTPConnectionPool|HTTP2|SwiftConfiguration': 167 XCTest + 32 Swift Testing configuration tests passed.
• Includes 15 deterministic state-machine regressions and a localhost TLS/HTTP2 test that verifies the public configuration opens two connections when each permits one stream.
• Regression tests were observed failing before the corresponding changes. Strict formatting and git diff --check passed.

Linux/minimum-toolchain validation, the full package suite, and throughput/latency benchmarks have not been run. The localhost test establishes concurrent capacity, not a measured performance improvement.

Does the default, required-event-loop exception, and demand-driven growth policy match the intended scope of #567?