fix(invariants): the email-handle floor is zero now, because they are all retired

[catomean]

The ratchet was set to 77 — a live population that could not simply be
renamed, because a username on OrangeCat is also a Lightning address and a
public profile URL.

They are now all retired, behind profile_username_history so nothing that
pointed at an old handle broke:

INSERT 0 53   history rows recorded first
UPDATE 53     handles replaced with user_<12 hex>
UPDATE 14     display names that were also the email local part cleared

Verified against production, not assumed

check	result
profiles where username = split_part(email,'@',1)	0
profiles where name = split_part(email,'@',1)	0
profile_username_history rows	54, none orphaned
count_email_derived_usernames()	0

And end to end on a real retired handle:

profile URL -> 308 https://orangecat.ch/profiles/user_a3eaa53c23cd
lnurlp      -> resolves the owner, answering under the NEW handle

That last detail is the proof: the LNURL endpoint replies about
user_a3eaa53c23cd when asked about the old handle, so the history lookup found
the right account.

Why 53 and not 77

Test-account churn between readings — the population moved 108 → 96 profiles
while E2E accounts were created and deleted, and 16 accounts had already signed
up under the fixed trigger. Every remaining one was renamed.

The constant stops being a ratchet

A violation now means a write path started minting them again. The three that
used to — the handle_new_user trigger, ensureProfile(), and two profile form
pre-fills — are each covered by a test, so a hit here means a fourth was
added. The message says that, instead of talking about a baseline that no longer
exists.

npm run verify green; 2385 tests pass.