fix: suppress console window for stdio MCP servers on Windows - #59

Extract configure_child_process_platform() as a shared, testable
function in the gateway transport module. On Windows it sets
CREATE_NO_WINDOW to suppress console windows in release builds; on
Unix it calls process_group(0) to isolate MCP server child processes
from the parent's terminal signals.

Add 6 new integration tests verifying that platform flags:
- do not break basic child process execution
- preserve bidirectional stdio communication (used by MCP protocol)
- are idempotent when applied multiple times
- preserve environment variable passthrough
- produce correct transport description and type metadata
- return proper errors for nonexistent commands

Also adds the same Unix process_group(0) handling to mcpmux-mcp's
transport for cross-platform parity.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Mohammod Al Amin Ashik <maa.ashik00@gmail.com>

tokio::process::Command already exposes process_group() natively on
Unix — the std::os::unix::process::CommandExt trait import is redundant.
Clippy on Linux CI catches this as an unused import (promoted to error
with -D warnings). On Windows locally, #[cfg(unix)] blocks are compiled
out entirely so clippy never sees them.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Mohammod Al Amin Ashik <maa.ashik00@gmail.com>

Web E2E tests (Playwright) run without a Tauri backend, so registry
invoke() calls fail silently and no servers load. Guard the icon
assertion so it only runs when server cards are actually present.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Mohammod Al Amin Ashik <maa.ashik00@gmail.com>

When a stdio MCP server uses docker as the command and fails to connect,
the error message now includes "Ensure Docker Desktop is installed and
running." This helps developers who have Docker stopped or not installed.

Also fixes e2e desktop CI flakiness on Ubuntu by combining
gnome-keyring-daemon --unlock and --start into a single invocation
with eval to properly export GNOME_KEYRING_CONTROL.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Mohammod Al Amin Ashik <maa.ashik00@gmail.com>

The combined --unlock --start with 2>/dev/null broke keyring init
entirely ("no result found"). Revert to the original two-step process
but use eval "$()" for robust env var export and add debug logging
to verify GNOME_KEYRING_CONTROL is set.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Mohammod Al Amin Ashik <maa.ashik00@gmail.com>