Fix disk-format handling and fail fast on host launch errors (#31/#32)

[Jordan231111]

Investigation

Refs #31 and #32 (the two newest open reports). The newer #33 is a closed provenance question without a runtime log and was also reviewed; no provenance/licensing material is changed.

Downloaded and read both complete public attachments:

• Cant install in android 11 #31: bsr_debug_20260907_204144.log — 11,590 bytes, SHA256 e1f88ad8b0eb4f2eb9cbe037230e4edd96ec25023d3d08cd700b5b069b35213f.
• acess denied eventhough WinDef had been remove #32: bsr_debug_20260913_124341.log — 3,060 bytes, SHA256 58638fe119ee6abf755eefadbe8cd6dfbe269f23a291a3ba70baa1c5e81573a2.

Both logs fail at line 28 while Windows tries to start HD-Player.exe, before guest ADB. #31 then polls for 120 seconds with no player or listener; #32's submitted log stops after about 17 seconds with the same absence. The #32 screenshot was also downloaded and reviewed: PREP finishes and detaches the updated disk, then the DATA-stage launch is denied. Removing Defender is not established as a solution; the logs do not identify the denying component.

Separately, #31's issue-body transcript reports a virtual-disk provider error during offline Root.vhd attachment. The code inferred storage type from the filename. Microsoft documents this behavior for Mount/Get/Dismount-DiskImage. A VHDX named .vhd is a reproducible failure case, but the reporter's actual disk format is NOT present in their log and is not assumed to be VHDX.

Changes

• Add canonical shared tools/bsr_host.ps1, embedded into both engines and debug.cmd by reembed.ps1.
• Detect VHD/VHDX by content; reject truncated, VDI, and unknown images rather than rename/convert them.
• Mount and query with an explicit storage type; detach using the successful mount's typed object. Do not detach pre-existing foreign mounts.
• Preserve the pristine-backup step, then preflight attachment before patching HD-Player or changing configuration.
• Use an explicit player working directory, quoted instance argument, and -ErrorAction Stop independent of the orchestrator's Continue preference.
• Preserve exception chains, native Win32 codes, HRESULTs, and native CIM error identifiers. Add read-only ACL/signature/policy-event observations to launch failures.
• Stop debug.cmd immediately with HOST_LAUNCH_FAILED and a nonzero status instead of polling an unstarted guest.
• Replace the misleading antivirus-blaming pipeline footer.
• Rebuild blueStackRoot.cmd; all non-script payload blocks remain byte-for-byte unchanged.
• Add host regression coverage to normal, read-only Windows CI.

Validation

Windows validation run 36308106993 passed all six suites: 430 checks passed, 0 failed, 2 existing environment-dependent skips. The new host suite passed 59 checks, including real disposable VHD/VHDX mounts, typed queries, and detach. The old extension-only path failed on a VHDX named .vhd with 0x80070570; the corrected path succeeded. This reproduces the code defect, not the reporter's exact disk or error code.

The two skips require ext4/debugfs tooling and an actual HD-Player executable. Temporary investigation workflows and source patchers have been removed. No release was published and main is unchanged.

Still requires reporter confirmation

Draft because the original Windows permission/policy denial has not been reproduced on the reporters' hosts. These changes fix the demonstrated code paths and diagnostics; they do not grant permissions or bypass host security policy. The existing logs are insufficient to distinguish ACL, AppLocker/App Control, or another host denial. No new ACL/security exclusions or policy changes are made by these helpers.

Please validate the updated diagnostic on Rvc64 (#31) and Tiramisu64 (#32). A remaining host denial should now include native codes and relevant observations; a remaining disk failure should identify the detected format and exact native error. Keep images and backups intact. Do not close either issue on CI results alone.