[BUG] fix HRPOpt.optimize raising AttributeError under scipy >= 1.18

[lenamonj]

Fixes item 3 of #750: HRPOpt.optimize() validates linkage_method against
scipy.cluster.hierarchy._LINKAGE_METHODS, a private attribute that scipy
1.18.0 removed, so every call raises AttributeError - including the default
linkage_method="single" - and CI is red on main (run 28899115146 at
a6638d2). pyproject.toml allows scipy>=1.3.0 unbounded, so a fresh
install resolves the broken combination.

Fix: delete the pre-check. scipy.cluster.hierarchy.linkage validates the
method itself and raises ValueError: Invalid method: <name> before touching
data, so the documented ValueError contract holds - the existing
test_hrp_errors assertion passes unchanged - and no private-API dependency
remains.

Test: test_hrp_linkage_methods runs optimize() over the seven linkage
methods documented by scipy.cluster.hierarchy.linkage and checks each
returns a full set of weights summing to 1. scipy exposes no public list of
valid methods, so the test pins the documented seven.

Measured on Python 3.13.8 / scipy 1.18.0 / numpy 2.5.1 / pandas 3.0.5 with
pytest ./tests -q:

• main: 7 failed, 295 passed, 15 skipped - all seven failures are this
  AttributeError (five in test_hrp.py, two in test_plotting.py, both of
  which construct an HRPOpt).
• The new test alone on main fails with the same AttributeError.
• This branch: 303 passed, 15 skipped, 0 failed. ruff check and
  ruff format clean on both changed files.

If you would like an upper bound (scipy>=1.3.0,<2.0.0) so a future scipy
2.0 cannot repeat this failure mode, happy to add it here or as a separate
PR, whichever fits your dependency policy.