Fix deadlocks, TCP stability, and ASCOM registration issues

[EricSchubert]

Please see attached markdown files for all detailed changes and fixes in this PR.
BUILD_FIXES.md
CHANGES.md

Summary

This PR fixes 11 bugs identified while using the ASCOM driver with N.I.N.A. and OATControl
on a BTT SKR Pico v1.0 running firmware V1.13.9. The issues range from permanent UI/thread
deadlocks to silent communication failures and COM registration problems.

Full details for each fix (problem, change, and rationale) are documented in CHANGES.md
and BUILD_FIXES.md, which I'll attach as comments below.

---

Runtime Bug Fixes (OATCommunications, ASCOM.Driver)

Fix 1 — Slew() deadlock (OatmealTelescopeCommandHandlers.cs)
doneEvent.Set() was missing from the :MS# callback — await doneEvent.WaitAsync() hung
forever. Added doneEvent.Set() inside the :MS# callback.

Fix 2 — SetLocation() deadlock (OatmealTelescopeCommandHandlers.cs)
A premature await doneEvent.WaitAsync() appeared before any commands were sent, making all
location/time commands unreachable. Removed the spurious wait.

Fix 3 — RefreshMountState() deadlock (OatmealTelescopeCommandHandlers.cs)
doneEvent.Set() was inside if (status.Success) — a comm failure meant it was never called
and the method hung. Moved Set() outside the if block.

Fix 4 — TCP timeout reported as success (TcpCommunicationHandler.cs)
A catch block substituted "0#" as a fake response, which CommandResponse treated as
success. Replaced with string.Empty and an explicit succeeded flag.

Fix 5 — DoubleFullResponse only reads once over TCP (TcpCommunicationHandler.cs)
DoubleFullResponse was grouped with single-response types — only one read happened, leaving
the second #-terminated reply in the buffer and corrupting the next command's response.
Split into its own case block with two reads, matching SerialCommunicationHandler.

Fix 6 — TCP stream leak on write failure (TcpCommunicationHandler.cs)
An early return on write failure bypassed stream.Close(), leaking the socket handle.
Added stream.Close() before the early return.

Fix 7 — TCP reconnects on every command (TcpCommunicationHandler.cs)
stream.Close() after every command closed the underlying socket, so _client.Connected was
always false and every command triggered a full TCP handshake. Added a _stream field to
keep the connection open; reconnects only when _stream == null.

Fix 8 — SendMessage() silently returns empty when disconnected (SharedResources.cs)
When SharedSerial.Connected was false, SendMessage() returned string.Empty —
indistinguishable from a legitimate empty response. Now throws ASCOM.NotConnectedException.

Fix 9 — PollUntilZero() hangs forever on mount stall (Driver.cs)
No timeout — a stalled mount or repeated "255" error returns from CommandString() caused
FindHome(), Park(), and synchronous slews to block the driver thread permanently. Added
maxAttempts = 120 (2-minute cap); throws TimeoutException on expiry.

Fix 10 — LocalServer32 registry path not quoted (LocalServer.cs)
Application.ExecutablePath was written without quotes. Paths with spaces caused
CO_E_SERVER_EXEC_FAILURE at COM activation. Wrapped in double quotes.

Fix 11 — SetupDialog() invisible when called from N.I.N.A. (Driver.cs)
COM-started processes are denied foreground privilege by Windows. SetupDialogForm opened but
was silently hidden behind other windows. Added SetForegroundWindow() on the main form
handle before ShowDialog() to regain foreground status.

---

Build / Project Fixes (OpenAstroTracker.csproj)

Build Fix 4 — Wrong NuGet HintPaths for ASCOM assemblies
HintPaths pointed to packages\ (inside the project dir) instead of ..\packages\ (the
actual NuGet restore location). ASCOM DLLs were not CopyLocal'd to bin/Debug/, causing
ReflectionTypeLoadException at COM activation on a clean build.

Build Fix 5 — Debug build targets wrong CPU architecture
Debug PropertyGroup had <PlatformTarget>AnyCPU</PlatformTarget>. On 64-bit Windows this
wrote the COM registration to the 64-bit registry hive; 32-bit ASCOM clients (N.I.N.A., SGP,
etc.) look in Wow6432Node and couldn't find the driver. Changed to x86 to match the
Release configuration and the official installer.

Note for maintainers: The Release configuration already uses x86 and is unaffected.
If your release pipeline generates Debug as an intermediate artifact, or if you prefer a
different mechanism for ensuring 32-bit registration in Debug builds, please review this
change. The core requirement is that LocalServer32 lands in Wow6432Node.

---

Test Environment

• Hardware: BTT SKR Pico v1.0, firmware V1.13.9
• ASCOM Platform 6.6, ASCOM driver v6.6.6.8
• N.I.N.A. 3.x (ASCOM telescope client)
• OATControl (WPF desktop app, WiFi + serial)
• Windows 11