refactor!: own the interpreter runtime toolchain rules instead of rules_python's

[jbedard]

Generated interpreter repos now instantiate a rules_py py_runtime_toolchain (py/private/interpreter/runtime.bzl, with a rules_py PyRuntimeInfo) instead of rules_python's py_runtime + py_runtime_pair + py_exec_tools_toolchain. One :runtime target per repo backs both toolchain registrations. Toolchain types are unchanged (@bazel_tools python type; rules_python's exec_tools type, with its exec_tools.exec_runtime/precompiler payload shape preserved), so rules_python-provisioned interpreters still satisfy whl_install/site-merge and rules_py still consumes user-registered rules_python py_runtime toolchains.

This shrinks each interpreter repo's rules_python loads from 31 .bzl files to the 12 of the deliberately kept py_cc_toolchain chain, insulating provisioned toolchains from rules_python implementation churn, and leaves the runtime layer pinning no rules_python minimum version.

New coverage: e2e/rules-python-interop (isolated module) pins that a module mixing both rulesets shares one rules_python-provisioned interpreter, including a $(PYTHON3) location diff_test; e2e/cases/rules-python-consumers pins that rules_python's own toolchain consumers (current_py_toolchain, exec-tools readers) work against rules_py-backed toolchains.

Changes are visible to end-users: yes

• Searched for relevant documentation and updated as needed: yes

• Breaking change (forces users to change their own code or config): yes

• Suggested release notes appear below: yes

• Generated interpreter repo targets are renamed: @python_X//:py3_runtime, :runtime_pair, and :exec_tools_toolchain are now a single @python_X//:runtime. Only direct label references need updating; toolchain resolution is unaffected.

• rules_py-provisioned runtimes carry rules_py's PyRuntimeInfo, not rules_python's. Provider-indexed access (target[PyRuntimeInfo] with rules_python's symbol) on these targets or resolved toolchains breaks; duck-typed field access keeps working. As a consequence, rules_python's py_binary/py_test cannot execute on rules_py-provisioned toolchains (they read provider fields rules_py does not carry) — the same one-way interop direction refactor: use simplified rules_py providers #1231 establishes for PyInfo.

• The exec-tools toolchain rules_py registers provides precompiler = None; enabling rules_python precompiling while @python_interpreters toolchains are registered first is unsupported.

Test plan

• Covered by existing test cases

[aspect-workflows]

✨ Aspect Workflows Tasks

📅 Thu Jul 16 08:30:48 UTC 2026

✅ 2 successful tasks

• ✅ test-e2e-bazel-9 [test] · ⏱ 2m 29s · 🐙 GitHub Actions
  💬 Bazel test complete (191/191 passed)
• ✅ test-examples-protobuf-bazel-9 [test] · ⏱ 2m 15s · 🐙 GitHub Actions · ☑️ Check
  💬 Bazel test complete (1/1 passed)

---

_{⏱ Last updated Thu Jul 16 08:33:12 UTC 2026 · 📊 GitHub API quota 427/15,000 (3% used, resets in 43m)
🚀 Powered by Aspect CLI (v2026.28.2)  |  Aspect Build · X · LinkedIn · YouTube}

[github-actions]

py_binary startup benchmark

Version	Mean (ms)	Median (ms)	± stddev	vs BCR	vs main	Build (s)
BCR 1.11.7 (baseline)	174.358	174.268	±0.886	—	—	35.23
HEAD main	56.026	55.808	±1.568	-67.9%	—	10.06
This PR	56.241	55.703	±2.076	-67.7%	+0.4%	8.28

Measured with hyperfine --warmup 5 --runs 50 on Linux
Gate: PR vs HEAD main (threshold: 10%). BCR is shown only as a historical baseline.
Build time: cold bazel build //:bench with isolated output base, no disk cache.

sys.path quality

Version	sys.path entries	distinct site-packages roots	duplicate realpaths
BCR 1.11.7 (baseline)	6	1	0
HEAD main	7	2	0
This PR	7	2	0

sys.path quality measured by bench_syspath inside the assembled venv. Duplicate realpaths indicate symlink redundancy; many distinct site-packages roots suggest an inefficient venv layout.

Bazel analysis benchmark

Version	Mean (ms)	Median (ms)	± stddev	vs BCR	vs main	Packages	Targets
BCR 2.0.0-alpha.4 (baseline)	10982.282	10955.149	±151.564	—	—	101	301
HEAD main	10458.960	10404.051	±240.645	-4.8%	—	101	301
This PR	10185.628	10192.962	±128.652	-7.3%	-2.6%	101	301

Measured with hyperfine --warmup 1 --runs 10 on Linux
Gate: PR vs HEAD main (threshold: 10%). BCR is shown only as a historical baseline.
Command: cold bazel build --nobuild //workspace/... with isolated output base, no disk cache.

Auxiliary metrics

Version	Loaded packages	Configured targets
BCR 2.0.0-alpha.4 (baseline)	101	301
HEAD main	101	301
This PR	101	301