Proof producer fails on balances_tree example

[color-typea]

Folder layout:

> ~/Projects/nil/zkllvm
> ~/Projects/nil/proof-market-toolchain

Steps to reproduce:

• Compile and assign balances_tree example in zkllvm: cmake --build build --target balances_tree_cpp_example_assign
• Prepare statement: python3 ./scripts/prepare_statement.py --circuit ../zkllvm/build/examples/cpp/balances_tree.cpp.ll --name balances_tree --type placeholder-zkllvm --private --output ./balances_tree_statement.json
• Run proof-producer: ./build/bin/proof-generator/proof-generator --proof_out=./balances_tree_proof.bin --circuit_input=./balances_tree_statement.json --public_input=../zkllvm/examples/inputs/balances_tree.inp

Expected result: produces the proof
Actual result: Errors with

./balances_tree_proof.bin_tmp.ll:58:48: warning: ptr type is only supported in -opaque-pointers mode
  %4 = getelementptr inbounds [6 x [2 x i64]], ptr @_ZZN3nil7crypto314multiprecision8backends11window_bitsEmE5wsize, i64 0, i64 %3
                                               ^
assigner: ./balances_tree_proof.bin_tmp.ll:58:48: error: expected type
  %4 = getelementptr inbounds [6 x [2 x i64]], ptr @_ZZN3nil7crypto314multiprecision8backends11window_bitsEmE5wsize, i64 0, i64 %3

Toolchain versions:

• Compiler: clang version 16.0.0 (git@github.com:NilFoundation/zkllvm-circifier.git f72301483aeb1d4857e621dd763092bbfab9a96f)
• Assigner: 0.1.1-5 (built from source on f72301483aeb1d4857e621dd763092bbfab9a96f)
• Proof-generator: built from source on latest master (commit 265097eb21fd0d64b6cb8fcf09f7efc32de6ca7c

Circuit, input, LL, etc.:

• Circuit: zkllvm/examples/cpp/balances_tree.cpp
• Input: zkllvm/examples/inputs/balances_tree.inp
• LL: see https://gist.github.com/color-typea/495faeeaa5ccc5ec0270d11c48bc83e7

Additional:
I'm not 100% sure, but it seems to be coming from __builtin_assigner_bit_decomposition64 builtin - offending code belongs to a

define dso_local noundef i64 @_ZN3nil7crypto314multiprecision8backends11window_bitsEm(i64 noundef %0) local_unnamed_addr #0 {

... which (if I read it right) is a function with a single int64 argument - and there are no such functions defined or used by the example except __builtin_assigner_bit_decomposition64