use wheel-backed scikit-learn on python 3.14

[sarptandoven]

summary

• route Python 3.14 installs to a scikit-learn range with native cp314 wheels
• keep Python <3.14 on the 1.6 series that current CI already resolves
• avoids the source build that made the Python 3.14 CI job several minutes slower

why

• issue investigate 3.14 test run time #412 tracks the Python 3.14 CI job taking roughly 6 minutes longer than the other supported Python versions
• recent CI runs show the delta is in the install step, not the test step:
  • run 28058718072: py3.14 install 263s vs py3.9-py3.13 installs 48-57s
  • run 28057448161: py3.14 install 256s vs py3.9-py3.13 installs 44-57s
  • run 28001330616: py3.14 install 266s vs py3.9-py3.13 installs 40-58s
• the Python 3.14 job resolved scikit-learn 1.6.1, which has no cp314 Linux wheel, so CI builds it from source
• scikit-learn 1.8.0+ publishes cp314 wheels, so Python 3.14 can install from wheels like the rest of the matrix

validation

• git diff --check
• poetry check
• verified no overlapping open PR for investigate 3.14 test run time #412 / Python 3.14 runtime / scikit-learn
• marker sanity check: Python 3.9-3.13 select >=1.6,<1.7; Python 3.14 selects >=1.8.0,<2.0
• pip download --only-binary=:all: --python-version 314 --implementation cp --abi cp314 --platform manylinux_2_28_x86_64 --no-deps 'scikit-learn==1.6.1' fails with no matching binary
• pip download --only-binary=:all: --python-version 314 --implementation cp --abi cp314 --platform manylinux_2_28_x86_64 --no-deps 'scikit-learn>=1.8.0,<2.0' resolves/downloads scikit-learn 1.9.0 cp314 manylinux wheel
• fresh Poetry 2.0.1 install on Python 3.11 resolves scikit-learn 1.6.1
• fresh Poetry 2.0.1 install on Python 3.14 resolves scikit-learn 1.9.0
• fresh clean clone from current upstream with only this diff applied:
  • make install on Python 3.14 installed scikit-learn 1.9.0
  • make lint passed
  • make test passed (86 passed, 2 xfailed, 22 xpassed)

closes #412

[alejandroschuler]

@ryan-wolbeck I have no idea what most of the words in the PR summary mean but the diff is small and just affects the versioning and you seem to think it's ok so going to approve :)