Add DPE ML-DSA support to Caliptra 2.0

[ArthurHeymans]

Squashed backport of:

• [dpe] Enable hybrid DPE feature ([dpe] Enable hybrid DPE feature #3243)
• [dpe] Add ML-DSA DPE Crypto trait ([dpe] Add ML-DSA DPE Crypto trait #3315)
• [dpe] Add support for 64 DPE contexts ([dpe] Add support for 64 DPE contexts #3246)
• [dpe] Add ML-DSA DPE command ([dpe] Add ML-DSA DPE command #3326)
• [dpe] Test ML-DSA CertifyKey ([dpe] Test ML-DSA CertifyKey #3357)
• [dpe] Test using external mu with ML-DSA profile ([dpe] Test using external mu with ML-DSA profile #3371)
• [dpe] reduce nesting of InvokeDpeCmd::execute ([dpe] reduce nesting of InvokeDpeCmd::execute #3386)
• [dpe] Add optional ML-DSA response over DMA ([dpe] Add optional ML-DSA response over DMA #3391)
• [dpe] Add HW model support for large DPE responses ([dpe] Add HW model support for large DPE responses in subsystem mode #3403)
• [dpe] Support both DPE profiles in more tests ([dpe] Support both DPE profiles in more tests #3407)
• [dpe] Add a test for worst case scenario certs and CSRs ([dpe] Add a test for worst case scenario certs and CSRs #3417)
• [dpe] Fail early for response sizing ([dpe] Fail early for response sizing #3415)
• [dpe] Add ML-DSA support to CertifyKeyExtended ([dpe] Add ML-DSA support to CertifyKeyExtended #3426)
• [dpe] Test ML-DSA profile with golang verification ([dpe] Test ML-DSA profile with golang verification #3454)
• [dpe] Add comment that initialization is profile agnostic ([dpe] Add comment that initialization is profile agnostic #3460)

DPE rev: a26db5b — last rev bumped by PRs in this list (#3371).

caliptra-2.0 adaptations:

• DpeMldsaCrypto uses MldsaReg (2.0 has dedicated ML-DSA HW register;
  main uses AbrReg shared Adams Bridge register for ML-DSA and ML-KEM)
• SignData::Mu returns CryptoError::NotImplemented (no external-mu on 2.0)
• Removed ML-KEM commands (no ML-KEM HW on 2.0)
• Removed main-only commands (ocp_lock, shake256)
• Flat PersistentData paths (2.0 uses single struct, not rom/fw split)
• Kept 2.0 caliptra-cfi-*-git package names to avoid ROM recompilation
• STACK_SIZE=136KB (main uses 154KB; 2.0 has ~30KB more PersistentData)
• RUNTIME_SIZE=176KB (main uses 210KB; 2.0 has fewer runtime features)

Additional fixes on top:

• Fix stack overflow in initialize_dpe by using execute_serialized
• Split InvokeDpe and CertifyKeyExtended into per-profile dispatch
• Defer common response buffer allocation to reduce stack usage
• Remove external AXI response support (unnecessary in 2.0)
• Remove subsystem mailbox size limit handling (mailbox not smaller in 2.0)
• Reduce MAX_HANDLES to preserve ROM layout
• Derive DPE context thresholds from MAX_HANDLES
• Fix tests for 32/64 context configurations