fix(org): match account names and logins case-insensitively

[vbhanuchander-lang]

Summary

ConnectUserAccountsExact links a user to an account through three exact comparisons: email,
FullName, and UserName. #9051 made the email comparison case-insensitive. The two name
comparisons were left matching exactly, and this makes them consistent with it.

The case that matters most is UserName, which holds the provider login. Provider logins are
themselves case-insensitive — GitHub resolves TyroneKCummings and tyronekcummings to the same
account — so comparing them case-sensitively against the users.csv name is comparing on a
distinction the provider does not make. Display names are folded for the same reason emails were: a
corporate git config and a provider profile routinely record the same name with different
capitalisation.

The consequence of a missed link is silent and easy to misread as missing data: the account is never
associated with the user, so every activity attributing through it disappears from work logs while
other activities for the same person keep appearing.

Does this close any open issues?

Towards #8698 — it does not close it. In that issue the GitHub account has no public email
(GitHub returns an empty string for users who have not made it public), so linkage can only happen
through the two name paths this PR fixes. That removes the case-mismatch failure, but a user whose
CSV name simply differs from their GitHub login is still unlinked. Closing #8698 properly needs a
way to declare the provider login rather than guess it, which is a larger change to the
crossdomain.User model and worth discussing on the issue first — user_account_mapping.csv
already covers the declared case and may be the preferred answer.

Other Information

Tested by extending the existing org e2e fixtures with account a12, whose login N7 differs from
user U007's name n7 only in case, and with no email so the match is forced down the login path.
With the change it links; reverting the change drops the expected user_accounts row count from 12
to 11 and the test fails.

gofmt and go vet clean. No documentation change: this corrects existing matching behaviour
rather than adding a knob.