Summary

Fixes #1433 by replacing the cryptic `pyparsing.ParseException` that users hit when declaring `_hidden: bool` with a clear `DataJointError` that explains the constraint and points to the alternative.

Design decision: don't allow user-defined hidden attributes

Earlier drafts of this PR loosened the parser to accept leading underscores, on the assumption that users should be able to declare hidden attributes the same way the framework does. On reflection, that's wrong:

• Hidden attributes are filtered out of every public API surface — `fetch()`, dict restrictions, joins, `insert`/`update1`, `describe()`. The filter is intentional and reflects the platform-bookkeeping intent.
• Platform-managed hidden columns (`_job_start_time`, `_job_duration`, `_job_version`, `_singleton`) are injected programmatically after parsing and populated via raw SQL during the `populate()` lifecycle (autopopulate.py). The parser never sees them.
• Allowing users to declare hidden attributes would expose a feature with no public-API write path, no `describe()` round-trip, and silent filtering on dict restrictions and `insert(ignore_extra_fields=True)`. That's a footgun.
• The cases users reach for hidden attributes — most commonly an index-backing derived column like `params_hash` — are better solved with a regular attribute. Backing a unique index isn't sufficient reason to hide a column. If application code computes, inserts, or queries the column, it should be a regular attribute.

So this PR keeps the parser strict and improves the error message instead.

Fix

Pre-flight check in `compile_attribute` (declare.py:858) that detects a leading underscore and raises a `DataJointError` before the parser is invoked:

```
Attribute name in line "_hidden: bool" starts with an underscore.
Names with leading underscore are reserved for platform-managed
columns (e.g. _job_start_time, _singleton). Use a regular attribute
name; if you need to control visibility at the call site, use proj().
```

The parser regex is unchanged from master. Platform code is unaffected because `job*` and `_singleton` bypass `compile_attribute` entirely.

Companion docs PR

datajoint/datajoint-docs#162 — reworked to reflect the new design. The §3.4 section is now framed as platform-only, the `_params_hash` user-defined example is removed, and the table of behaviors is preserved as a reference for the platform-managed columns.

Test plan

• `tests/unit/test_declare_hidden_attribute.py` — 6 tests: `compile_attribute` rejects `_hidden`, `params_hash`, and leading-whitespace variants with the helpful message; parser still rejects leading ``; parser still accepts plain names; parser still rejects digit-start.
• Full unit test suite (255 tests) — all pass, no regressions.
• `ruff check` and `ruff format` clean.