1.8: fix(ohcldiag-dev): treat GUID-shaped VM args as names, not IDs (#4042)

[moor-coding]

Backport of #4042 to release/1.8.2607.

Why this is needed for 1.8

This regression (introduced by #3753, which is present on release/1.8.2607) breaks all ohcldiag-dev interactions against fleet VMs — it is currently blocking Underhill 1.8 memory validation on TiP.

Problem

#3753 changed VmId::from_str to try parse::<guid::Guid>() first for bare and hyperv:-prefixed arguments, treating any successful parse as a VM ID. In the fleet, Hyper-V VM names are themselves GUID-shaped, so a name is misinterpreted as a VM ID and used directly as the AF_HYPERV connect address — instead of being resolved to the real VM ID via hvc.exe id <name>. The bogus address produces:

Error: failed to connect
Caused by:
    0: failed to connect
    1: The requested address is not valid in its context. (os error 10049)   // WSAEADDRNOTAVAIL

Fix

Make ID-vs-name resolution deterministic instead of inferring intent from the string's shape:

• hyperv-id:GUID → VM ID (new, explicit prefix — the only way to look up by ID)
• hyperv:NAME → name (even if GUID-shaped)
• bare value → name (unless it is a unix socket path → hybrid vsock)
• vsock:PATH → hybrid vsock (unchanged)

Also change FromStr::Err from Infallible to ParseVmIdError so an explicit hyperv-id: with an invalid GUID reports a clear parse error.

Behavior change

Callers relying on a bare GUID being interpreted as a VM ID (introduced in #3753) must now use the hyperv-id: prefix.

Testing

• Clean cherry-pick of the main fix (no conflicts).
• Added Windows-gated unit tests for VmId::from_str covering each parse path.
• cargo check (Linux) + cargo clippy --all-targets / cargo doc --no-deps (x86_64-pc-windows-msvc) clean against the 1.8 toolchain (1.95.0).

Depends on #4042 (original merge to main).

[Copilot]

Pull request overview

This PR backports the ohcldiag-dev VM-argument parsing fix to release/1.8.2607, making Hyper-V VM lookup deterministic so GUID-shaped fleet VM names are no longer misinterpreted as VM IDs (the regression introduced by attempting GUID parsing first).

Changes:

• Introduces an explicit hyperv-id:GUID prefix to opt into direct Hyper-V VM-ID addressing; hyperv: and bare values are treated as VM names (even if GUID-shaped).
• Updates VmId::FromStr to emit a concrete parse error (ParseVmIdError) for invalid hyperv-id: GUIDs.
• Adds Windows-gated unit tests covering the intended parse behaviors.

Comments suppressed due to low confidence (1)

openhcl/ohcldiag-dev/src/main.rs:390

• ParseVmIdError currently has all variants gated with #[cfg(windows)], so on non-Windows it becomes an empty enum while still deriving thiserror::Error. Gating the whole type to Windows (and removing the per-variant cfg) avoids the empty-enum edge case and makes the cfg intent explicit.

/// Error parsing a [`VmId`].
#[derive(Debug, Error)]
enum ParseVmIdError {
    #[cfg(windows)]
    #[error("invalid VM ID GUID '{0}' (expected a GUID, with or without braces)")]