rules_python 1.6 cannot find python 3 (local toolchain)

# 🐞 bug report

### Affected Rule

`rules_python/tools/wheelmaker`

### Is this a regression?

We did not encounter this in 1.4.2: https://github.com/protocolbuffers/protobuf/actions/runs/17471667264/job/49621377398

### Description

I upgraded protobuf's rules_python version to 1.6 from 1.4.2. I'm now running into an issue when trying to mirror module.bazel deltas to WORKSPACE.

It looks like rules_python is having difficulty locating python3 for localtoolchains after updating to 1.6.

## 🔬 Minimal Reproduction

https://github.com/protocolbuffers/protobuf/pull/23341

## 🔥 Exception or Error

```
File "/workspace/_build/out/sandbox/processwrapper-sandbox/1203/execroot/com_google_protobuf/bazel-out/k8-opt-exec-ST-a828a81199fe/bin/external/rules_python/tools/wheelmaker", line 411, in _RunExecv
os.execv(python_program, argv)
FileNotFoundError: [Errno 2] No such file or directory: '/usr/local/bin/python3'
INFO: From Compiling upb_generator/reflection/header.cc [for tool]:
In file included from upb_generator/reflection/header.cc:15:
In file included from bazel-out/k8-opt-exec-ST-a828a81199fe/bin/src/google/protobuf/compiler/_virtual_includes/code_generator/google/protobuf/compiler/code_generator.h:2
```

## 🌍 Your Environment

**Operating System:**

Linux

**Output of `bazel version`:**

bzl 7.*, 8.*

**Rules_python version:**
1.6.0

Passing all but 1 orthogonal test in 1.4.2: https://github.com/protocolbuffers/protobuf/actions/runs/17471667264/job/49621377398

Failing a lot of Linux in 1.6: https://github.com/protocolbuffers/protobuf/actions/runs/17469109780/job/49612587161?pr=23341

Commencez par rules_python/tools/wheelmaker et reproduisez l’échec en utilisant la PR protobuf liée dans l’issue, en comparant rules_python 1.4.2 à 1.6.0 ainsi que la configuration module.bazel-to-WORKSPACE. Vérifiez les échecs de la CI Linux et le chemin de la toolchain locale affiché dans le traceback ; le travail est terminé lorsque les builds Linux concernés trouvent Python 3 et que la régression est couverte par des tests réussis.

Rédigé par le modèle d'indexation à partir du texte de l'issue.