fix(bundler): treat an explicit-null records field as missing, not the text "None"

[jawwad-ali]

Problem

InstalledBundleRecord.from_dict and _component_from_dict read the required fields with:

str(data.get("bundle_id", "")).strip()

The "" default covers only a missing key. A key that is present but null — how a hand-edited or corrupt .specify/bundle-records.json spells an empty field — yields None, and str(None) is the literal "None". That text is non-empty, so it sails straight past the if not bundle_id: / if not version: / if not cid: guards that exist precisely to reject such a file.

Reproduction on current main (bf88c9f)

bundle_id=null -> ACCEPTED: bundle_id='None' version='1.0.0'
version=null   -> ACCEPTED: bundle_id='a'    version='None'
id=null        -> ACCEPTED: components=[('presets', 'None')]

The record is accepted as a bundle literally named None at version None, contributing a component whose id is None.

That phantom is not inert: it feeds the collateral-protection refcount, so components_still_needed() reports ('presets', 'None') as protected, and install_bundle()'s other_tracked set will attribute real components to it.

Fix

Reuse _text from the sibling parser in the same package. Its docstring describes this exact trap:

A .get(key, "") default only covers a missing key. A key that is present but null — how YAML spells an empty field (author: with nothing after it) — yields None, and str(None) is the literal "None". That text is non-empty, so it sailed past the if not value required-field checks…

This is the second half of a hardening sweep that was already applied here in part: the falsy-non-list guards landed in this file as #3666, and the explicit-null fix landed for bundle.yml as #3798 — but was never mirrored into records.py. Using the shared helper keeps the two parsers from drifting again.

No breaking change. Only null-valued inputs behave differently; every string input is byte-identical. installed_at is optional, so its null correctly becomes "" rather than an error — pinned by a test.

Verification

• Fail-before / pass-after: 4 new-vs-baseline failures with the source reverted, all passing with the fix — 5 failed → 1 failed (that one is a pre-existing Windows symlink test).
• Normal records still load unchanged: bundle_id='demo' version='1.0.0' comps=[('presets','p1')].
• Scoped regression over tests/unit: no new failures vs a clean-main baseline captured on bf88c9f9.
• uvx ruff@0.15.0 check src tests → clean

Tests sit beside the existing test_load_records_rejects_record_missing_bundle_id / _missing_version / _rejects_component_missing_id, which cover the missing-key half of the same contract.

---

Written with assistance from Claude Code. Bug found, reproduced, and verified by me on current main.