Sharpen EvidenceItemSupportEnum descriptions in schema (#56)

* Sharpen EvidenceItemSupportEnum descriptions in schema

The previous one-liner enum descriptions ("Statement is incorrect",
"Evidence partially supports the claim", etc.) didn't distinguish
between REFUTE and WRONG_STATEMENT, and didn't say what makes a claim
"partial" versus fully supported. Curators were using WRONG_STATEMENT
for misattributed references (paper doesn't address the claim) while
REFUTE is meant for references that actively contradict the claim;
the schema now spells out that contrast.

- SUPPORT: snippet substring + surrounding context endorses the
  claim.
- REFUTE: reference actively contradicts (opposite effect, mechanism
  not occurring, etc.). Currently unused, kept for future curation.
- PARTIAL: supports part but not all - common cases listed.
- NO_EVIDENCE: placeholder hypothesis without supporting reference.
- WRONG_STATEMENT: reference is misattributed - paper does not
  address the claim at all (wrong DOI/PMID, different system).
  Explicitly contrasted with REFUTE.

Also includes:
- Regenerated `src/communitymech/datamodel/communitymech.py` from
  the schema (just gen-python). Most of the diff is generator
  version drift (single-quote -> double-quote, import
  consolidation); the semantic change is the new enum docstrings.
- Incidental black auto-formatting of literature.py and two test
  files that were already showing as "would be reformatted" on main
  before this PR. `just format` picks up the entire tree, so these
  came along for free; reverting them would just re-create the
  same lint warning.

Test plan:
- `just validate-all`: passes (no issues found across 255
  communities).
- `just test`: 102 passed, 9 skipped.
- `just lint`: black is now clean; ruff still has 317 pre-existing
  F401/etc. errors unrelated to this PR.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Address Copilot review on PR #56

- NO_EVIDENCE description rewritten so it aligns with the schema's
  requirement that EvidenceItem.reference and EvidenceItem.snippet
  are present: the enum now means "cited paper is on-topic but the
  curator could not locate a supporting or contradicting excerpt;
  the snippet records the closest on-topic excerpt examined."
  Explicitly contrasted with WRONG_STATEMENT (misattribution) and
  REFUTE (active contradiction) so curators have a clear rule.
- Exclude the LinkML pythongen datamodel from ruff. The file is
  regenerated wholesale by `just gen-python` and not hand-edited, so
  E501/UP045/etc. lints on the generated output produce churn rather
  than signal. black still formats it via `just format`. Cuts
  `just lint` output from 317 -> 174 ruff errors.
- Regenerated datamodel pulls in the updated NO_EVIDENCE docstring.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: realmarcin

pyproject.toml

@@ -68,6 +68,10 @@ target-version = ["py310", "py311", "py312"]
 [tool.ruff]
 line-length = 100
 target-version = "py310"
+# The LinkML pythongen datamodel is fully regenerated by `just gen-python`
+# and is not hand-edited, so style lints there create churn rather than
+# signal. Exclude it from ruff (black still formats it via `just format`).
+extend-exclude = ["src/communitymech/datamodel/communitymech.py"]
 
 [tool.ruff.lint]
 select = ["E", "F", "I", "UP", "N", "S", "B", "A", "C4", "T20", "SIM"]