fix: accept option-like updater helper values

[eWOOD29]

Scope

Narrow follow-up to the accepted PR #5 integration. develop merge commit f5ffebf1f1643ed326cb44910778b89fe5d06c9d exposed an updater-helper CLI compatibility edge case in Windows 3.13 CI: the published v0.2.1 parent passes the random helper handshake value as a separate argv entry, and a value beginning with - can be interpreted by argparse as another option before the staged helper reaches preflight.

This draft changes only the staged helper compatibility boundary and adds focused regressions. It does not change Stable main, publish a Beta, or modify the live installation.

Fix

• normalize only the legacy split helper handshake argument when its value is option-like, syntactically valid, and not itself a known helper option;
• preserve the published v0.2.1 parent invocation without requiring a stable-parent code change;
• keep ordinary helper parsing and all existing preflight/update/restart/rollback behavior unchanged;
• add regressions proving an option-like value is bound before argparse and a known option is not accidentally consumed.

Frozen evidence

• Base/develop: f5ffebf1f1643ed326cb44910778b89fe5d06c9d
• Head: 640eaf75cb0b9e09c62a73f1fbade1be1862fe26
• Head tree: 51d2c4539bf2e9902c9d5e271645b9d3da5fea6e
• Ancestry: 2 commits ahead / 0 behind
• Changed files: exactly scripts/update_helper.py and tests/test_update_helper_cli_regression.py
• Exact-head CI: 31754027579 — SUCCESS
• Windows Python 3.11/3.13: 203 tests, zero skips
• Ubuntu matrices, Windows portable install/uninstall, portable package equality, private fixtures, private preview/digest/ZIP comparison, JavaScript, compile, privacy, and docs: PASS

PR #6 remains intentionally draft. Fresh focused independent review of this exact head is required before any merge into develop. Beta publication and live Beta opt-in remain separate gates.