[do-not-merge] add Lattigo backend support and end-to-end CKKS execution fixes

[copybara-service]

[do-not-merge] add Lattigo backend support and end-to-end CKKS execution fixes

Introduce backend emitter features, parameter generation adjustments, and
pipeline controls required for compiling and executing deep CKKS Transformer
models (e.g. BERT-tiny SST-2) on the Lattigo backend:

• Strided subview lowering & memref copy codegen:

  • Support non-contiguous memref.subview operations resulting from tensor.pad
    bufferization by aliasing the whole source slice in Go and tracking
    aliases.
  • Lower memref.copy between strided and contiguous views into nested Go loops
    with explicit flattened multi-dimensional affine index arithmetic.

• Wide-message bootstrapping & declared-scale bundles:

  • Introduce --lattigo-bootstrap-declared-scale-log2 in LattigoEmitter to wrap
    emitted bootstraps in declared-scale bundles for wide-range messages.
  • Temporarily declare input ciphertext scale as Delta * 2^s so Lattigo's
    ScaleDown brings the message into the Mod1 Chebyshev approximation domain
    (|m| <= 2^s).
  • Restore the true message scale post-bootstrap via exact integer scalar
    multiplication by 2^s without consuming ciphertext levels or rescaling.
  • Cache module parameter lookups in LattigoEmitter to validate first modulus
    bit-depth and eliminate redundant AST walks.

• 61-bit prime generation & deep-circuit RNS decomposition:

  • Implement lastPrimeCapAware with a 64-bit deterministic Miller-Rabin test
    in CKKS/Params.cpp to find first primes up to 61 bits congruent to 1 mod
    2N, bypassing OpenFHE's 60-bit MAX_MODULUS_SIZE limit.
  • Update computeDnum in RLWEParams.cpp to select dnum=4 for deep circuits
    (level >= 16), sizing auxiliary P primes compactly to satisfy 128-bit
    quantum security standards.

• Trivial zero plaintext sizing:

  • In ImplementTrivialEncryptionAsAddition, size trivially-encrypted zero
    constants using the physical slot count (scheme.actual_slot_count) instead
    of the logical requested slot count.
  • Prevent sparse zero plaintexts on rings expanded for security constraints.

• Pipeline & build configuration:

  • Add alloc-to-in-place option to ToLattigoPipelineOptions in
    ArithmeticPipelineRegistration, enabling circuits with complex shift
    networks to bypass in-place reuse hazards.
  • Declare ctx.executable._heir_translate_binary in tools attribute of
    heir-translate.bzl for correct Bazel tool tracking.

• Testing:

  • Add lit tests for strided memref.subview copies with nested Go loops in
    memref.mlir.
  • Add lit tests validating default bootstrap and declared-scale bundle
    generation under --lattigo-bootstrap-declared-scale-log2 in
    bootstrap.mlir.
  • Add regression lit test validating zero plaintext sizing against physical
    actual_slot_count in
    implement_trivial_encryption_as_addition_actual_slot_count.mlir.