Drop deprecated copy keyword from astype call in from_fcs

[xyf5432]

PR: Guard deprecated astype copy keyword by pandas version in from_fcs

Fixes #183

Changes

Guard the deprecated copy keyword of DataFrame.astype by pandas
version in from_fcs, so behaviour is preserved on every supported
version:

• src/palantir/io.py: module-level PANDAS_VERSION = Version(pd.__version__)
  (from packaging, already a transitive dependency of pandas), and

  if PANDAS_VERSION < Version("3.0"):
      data = data.astype(np.float64, copy=False)
  else:
      data = data.astype(np.float64)

  — the same pattern as Fix pandas 3 string dtype compatibility and CI flakes NVIDIA/cuml#8142.

• README.md: changelog entry under Version 1.4.5, alongside the
  pandas 3 fixes from Pandas 3 compatibility: ValueError: assignment destination is read-only in _differentiation_entropy #180.

Verification

The copy keyword never affects the result of astype on any
pandas version, only whether an extra copy is made; the branch keeps
that behaviour exactly as it was on each side of the boundary. Runtime
checks on pandas 2.3.3 and 3.0.5, simulating the from_fcs data flow
(fcsparser.parse-like DataFrame → astype → metadata-channel filter →
np.arcsinh(np.divide(...))):

• pandas 2.3.3 takes the copy=False branch: already-float64 columns
  still share the buffer (np.shares_memory = True), exactly as before
  the change; int64 → float64 columns allocate a new array either
  way (a dtype change cannot be done in place).
• pandas 3.0.5 takes the plain branch: no Pandas4Warning is emitted,
  final arcsinh results bit-identical to pandas 2.3.3.
• Modified files pass python -m py_compile.

[katosh]

Hi @xyf5432,

Thank you for the issue and PR! This overall looks great, there is just two things we should do:

1. v1.4.5 already shipped. Write a new ### Version 1.4.6 (future) section in the changelog instead.
2. Since copy=False can only skip an allocation when the frame is already float64, the old code was inert, and we can simply do data = data.astype(np.float64) and skip the version check. This also avoids introducing the packaging dependency.

Let me know if you need help with that!