fix(specs): unwrap ju rjsf form spec + modernize packaging metadata

[thorwhalen]

The bug

uf/specs.py called func_to_form_spec(func, **self._rjsf_config). But
ju.rjsf.func_to_form_spec defaults to nest_under_field="rjsf", so it returns:

{"rjsf": {"schema": {...}, "uiSchema": {...}}}

_generate_spec then read form_spec.get("schema", {}) — which always yielded
{}. Every generated RJSF form was empty whenever ju imported
successfully. This is user-facing, not a test artifact: uf/routes.py serves
spec["schema"] straight to the browser in two places.

This was masked in CI because the broad except ImportError: around the ju
call block silently falls back to _generate_basic_spec, which happens to emit
the flat shape — so CI was green while the ju integration (the package's whole
point) was never actually exercised.

The fix

Pass nest_under_field=None, via a dict merge so caller overrides in
_rjsf_config still win:

rjsf_config = {"nest_under_field": None, **self._rjsf_config}
form_spec = func_to_form_spec(func, **rjsf_config)

Safe in both environments: it makes the ju path emit exactly the same flat
shape the ImportError fallback already emits. The except ImportError: clause
is deliberately left untouched in this PR (see Deliberately out of scope).

Test counts

	before	after
python -m pytest -q	2 failed / 27 passed	29 passed
dependents gate (test-dependents uf)	0 pass / 1 fail	1 pass / 0 fail

Previously failing: tests/test_specs.py::test_function_spec_schema_basic
(KeyError: 'type') and tests/test_specs.py::test_function_spec_required_params
(assert 'name' in []).

Packaging metadata

• license: deprecated [project.license] text = "mit" table -> PEP 639 SPDX
  string license = "MIT" (canonical case). No License :: OSI Approved ::
  trove classifier added — PyPI rejects that combination alongside an SPDX
  expression.
• keywords / authors: were empty placeholders ([]), now filled in.
• classifiers: were absent entirely, so the PyPI page carried no trove
  metadata. Eight added.
• description: "UI Fast" -> the README tagline
  "Minimal-boilerplate web UIs for Python functions". The GitHub repo
  description is set to the same string so the two do not drift.
• urls: added Documentation = "https://i2mint.github.io/uf/" (already live
  on the gh-pages branch).
• .editorconfig: new, copied verbatim from the fleet reference.

Verified with uv build: the wheel METADATA emits License-Expression: MIT
plus the eight classifiers, with no conflicting license classifier. ruff check
and ruff format --check both clean.

Repo settings were also updated out-of-band (not branch-scoped): homepage set to
the docs URL, and topics added.

Deliberately out of scope

Left for later specialist passes, each a known finding:

• The broad except ImportError: in uf/specs.py — currently load-bearing
  for CI green. Something in the ju import chain raises ImportError in the
  GitHub runner environment (not yet identified; ipywidgets, lazy graphviz,
  and every ju/__init__ name were ruled out against the published wheel).
  Narrowing it now would surface that hidden ImportError as a CI failure. It
  deserves its own investigation.
• 38 of 76 package doctests fail — illustrative Example: blocks written in
  doctest syntax that were never runnable. testpaths = ["tests"] is the only
  thing keeping them out of collection, so testpaths is intentionally
  unchanged here; widening it before those are fixed would turn CI red.
• No uf/py.typed, no CHANGELOG, CI is still uv-inline rather than the
  reusable-workflow stub.

Upstream note (informational, not uf's job): func_to_form_spec nesting its
result under "rjsf" by default is an easy trap for every consumer of ju.

https://claude.ai/code/session_01VipiLaG4xy7WctqY9w2475