Fixes #4383

Problem

/speckit.tasks maps data-model.md entities to user stories (templates/commands/tasks.md, "From Data Model" section) but never instructed the agent to carry field-level constraints (max length, nullable/required, enum values, validation rules) into the generated task text. Left to the agent's discretion, a constraint present in data-model.md can be silently dropped from tasks.md, and the implementing agent then invents its own value (e.g. the reporter's data-model.md said SuspendReason nvarchar(200) but the generated code used SuspendReasonMaxLength = 500 because tasks.md never repeated the 200 constraint for that field).

Fix

Added one bullet to the "From Data Model" task-organization rule in templates/commands/tasks.md requiring that field constraints from data-model.md be quoted verbatim in the corresponding task description, so the value isn't left to implementation-time discretion.

This is a prompt-only change — no Python code, no breaking changes.

Testing

Added tests/test_tasks_template_constraints.py, which asserts the "From Data Model" section of templates/commands/tasks.md mentions carrying constraints forward.

Verified the test fails without the fix and passes with it:

$ git checkout HEAD~1 -- templates/commands/tasks.md   # (unfixed content)
$ python -m pytest tests/test_tasks_template_constraints.py -v
FAILED tests/test_tasks_template_constraints.py::test_data_model_section_requires_verbatim_field_constraints
AssertionError: The 'From Data Model' task-organization rules must instruct the agent to carry
field constraints (max length, nullable, enum, validation rules) from data-model.md into task
descriptions.
1 failed in 0.49s

$ git checkout HEAD -- templates/commands/tasks.md      # (fixed content)
$ python -m pytest tests/test_tasks_template_constraints.py -v
tests/test_tasks_template_constraints.py::test_data_model_section_requires_verbatim_field_constraints PASSED
1 passed in 0.17s

Full suite:

$ python -m pytest tests -q --ignore=tests/hooks
10 failed, 7691 passed, 12 skipped, 48 warnings in 473.70s

The 10 failures are pre-existing and unrelated to this change — they are all *_python_parity tests exercising the "composed" preset variant (test_check_prerequisites_python_parity.py, test_create_new_feature_python_parity.py, test_resolve_template_python_parity.py, test_setup_plan_python_parity.py, test_setup_tasks_python_parity.py). I confirmed the identical 10 tests fail on a clean checkout of upstream/main with no changes applied, in this sandbox (likely missing an environment prerequisite for the composed-preset fixtures, e.g. PowerShell). All tests in the affected files pass individually in isolation.

AI Disclosure

I did use AI assistance (Claude, via Claude Code) to investigate the issue, write the template change, and write/verify the regression test.