fix(deps): apply security fixes to td-shim-AzCVMEmu

[MichalTarnacki]

#	Assessed Severity	Bug	Call Chain / Location	Fix	Classification	Disposition Reason
1	Low	tdcall_servtd_rebind_approve indexes target_td_uuid/token without length check	-> tdcall_servtd_rebind_approve	Use .get(..) and return Err on short slices.	weakness	The defect is real; the emulated tdcall_servtd_rebind_approve indexes target_td_uuid[0..3] and run copy_from_slice(&rebind_session_token[..32]) with no length check, so a slice shorter than 4 u64 or 32 bytes panics (CWE-248). But it is not an exploitable true positive. This is AzCVMEmu emulation code, outside the production TDX TCB — in the shipped product this call is a hardware TDCALL, and the production wrapper in tdx.rs:952-958 already validates both lengths (target_td_uuid.len() != 4 || rebind_session_token.len() != REBIND_SESSION_TOKEN_SIZE) and returns Err before any indexing.
2	Medium	free_shared_pages: Box::from_raw with caller-supplied num -- Layout mismatch + ov	-> free_shared_pages	Store original Layout alongside allocation; dealloc with that. Return Err on overflow.	weakness	Split the finding into its two claims. The "panic on huge num" via checked_mul(num).expect() is a false positive: num is never attacker-chosen. It is derived internally as dma_size / PAGE_SIZE, and dma_size = align_up(buf.len()) where buf.len() is hard-capped at MAX_VSOCK_PKT_DATA_LEN (64 KiB) by the enqueue guard, so num ≤ 16 and 0x1000 * 16 cannot overflow usize. The huge-num path is mathematically unreachable. The "Layout mismatch → heap UB" half is a real defect but only a weakness, not the T1/MUST-FIX it is labeled. Reconstructing the dealloc Box from the caller's num instead of the recorded allocation size is a genuine footgun in an unsafe fn, but it only bites if some caller frees with a different count than it allocated. No in-tree caller does that — every path (the vsock Allocator, and virtio_pci storing dma_size and freeing with dma_size / PAGE_SIZE) is self-consistent. Crucially, the VMM does not control the free num; it controls bounded packet contents/lengths, while the page count is internal allocation bookkeeping. So the "directly reachable from untrusted VMM / oob_write" tag is overstated. This is also emulation-only code (td-payload-emu), outside the production TDX TCB.
3	Low	register_interrupt_callback truncates usize vector to u8	-> register_interrupt_callback	Change parameter type to u8 or assert vector<256.	weakness	The vector parameter is always supplied by MigTD's own code (hardcoded interrupt vector constants passed to register_interrupt_callback), not by any external or VMM-controlled input. The truncation from usize to u8 is only dangerous if a caller passes a value ≥ 256, which no current caller does. The guard converts a hypothetical future misuse into a clean error return rather than silent index wraparound, but today the dangerous path is unreachable — making this defense-in-depth hardening, not remediation of an exploitable vulnerability.