feat(text): _Paragraph.fields field-discovery accessor (Phase 4)

[MHoroszowski]

Phase 4 of #20 — field discovery accessor

The small read-side companion to Phase 3's authoring API (PR #50). Adds _Paragraph.fields, a parallel to _Paragraph.runs that returns the paragraph's <a:fld> children wrapped as _Field instances in document order. Combined with Phase 3, the round-trip story closes:

# Phase 3: author
p = slide.placeholders[1].text_frame.paragraphs[0]
fld = p.add_field()
fld.type = "slidenum"
prs.save("out.pptx")

# Phase 4: discover and mutate
prs = Presentation("out.pptx")
p = prs.slides[0].placeholders[1].text_frame.paragraphs[0]
for f in p.fields:
    if f.type == "slidenum":
        f.text = "Page <#>"

Why this PR is small

The heavy lift landed earlier:

• Phase 1 (PR feat(oxml): CT_HeaderFooter + CT_HandoutMaster + CT_TextField attrs (headers-footers Phase 1) #48) — CT_TextField element class with id/type attrs.
• Phase 3 (PR feat(text): Field authoring — _Paragraph.add_field, _Field class, CT_TextField.text setter (Phase 3) #50) — _Field class, CT_TextParagraph.fld ZeroOrMore (gave us fld_lst for free), field-aware content_children and _Paragraph.text.

Phase 4 is one @property body — wrap self._element.fld_lst in _Field instances and return as a tuple. The existing fld_lst does all the work.

Anti-criteria upheld

• _Paragraph.runs continues to yield only _Run (regression-pinned by it_keeps_runs_field_free_on_mixed_paragraphs).
• _Paragraph.text semantics unchanged (already field-inclusive since Phase 3).
• _Field class itself unchanged — read-only addition.

Verification (local, CPython 3.14.4)

python3 -m pytest tests/ -q                       → 3632 passed in 5.24s (+6 vs Phase 3 baseline)
python3 -m ruff check src tests                   → All checks passed!
python3 -m ruff format --check src tests          → 216 files already formatted
python3 -m behave features/ --no-color            → 1048 scenarios, 0 failed
python3 uat/uat_headers_footers_phase4.py         → PASS

UAT confirms the full Phase 3 → Phase 4 round-trip: opens uat/out_headers_footers_phase3.pptx, walks to the slidenum field via .fields, mutates its text via the discovered _Field handle, re-saves, re-opens, and reads it back via .fields again — clean round-trip.

6 new tests in tests/text/test_text.py Describe_Paragraph:

• it_returns_an_empty_tuple_of_fields_when_paragraph_has_none
• it_provides_access_to_a_single_field
• it_yields_multiple_fields_in_document_order
• it_chains_each_field_parent_back_to_the_paragraph
• it_returns_a_tuple_not_a_list
• it_keeps_runs_field_free_on_mixed_paragraphs (anti-regression)

What's next

Phase 5 closes out the epic with the HandoutMaster Python class + HandoutMasterPart plumbing, and the watermark helper. After that, issue #20 ships.

Refs #20.