Say what the login form layout does, not what it replaced

[JeroenDeDauw]

Follows-up to #40

Three comments and a test name described the login form as it was before that pull request rather
than as it is: the watermark "led the form while nothing was described below zero", core's button
had to be demoted "now that" the form recommends the member route, and a test was named for that
button no longer being primary. A reader of master cannot check any of it, and the mechanism that
makes the code necessary — an unweighted field sinks under a section weighted below zero — was left
implied. Each now states that mechanism in the present tense.

Two smaller ones. PASSWORD_FIELD and PASSWORD_LOGIN_BUTTON shared one gloss though they answer
different questions, and they disagree on a wiki whose only other route is single sign-on, where the
divider is suppressed while the demotion still fires. And LoginCodeRequest asserts that the form's
first button names no route without pointing at what makes that true.

Comments only: the code tokens of both source files are unchanged and the stylesheet is untouched,
so the rendered form cannot differ.

Considered, omitted:

• Deleting testDefaultButtonIsLeftOutOfTheTabOrder, which no mutation reddens on its own. It names
  a promise a reader would look for, and dropping a test to save three lines is a poor trade.
• Folding the mfLogo re-weight into CAPTCHA_WEIGHTS, or a shared helper for the two: two call
  sites with different guards and different reasons.
• Reusing fallbackLine() for the hidden submit field: would add a parameter for one caller and
  rename the concept.
• Renaming testMobileWatermarkGoesOnLeadingTheForm: "goes on" reads as the invariant the handler
  preserves, matching testCollectedFieldStaysBelowTheButtonThatEntersTheCode beside it.
• The five negative weight constants, the single-assignment private method, and the field-name
  constants: each follows a convention the class already had.

_{AI-authored — Claude Code, Opus 5 (max); simplify skill run on a detailed brief from @JeroenDeDauw, no course corrections; diff not yet human-reviewed; review run as four subagents, the change proven comment-only by comparing PHP code tokens, phpcs, phpstan and PHPUnit green locally, and the renamed test checked against a mutation of the code it covers.}

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.71%. Comparing base (e53204d) to head (9772a66).

Additional details and impacted files

@@            Coverage Diff            @@
##             master      #41   +/-   ##
=========================================
  Coverage     87.71%   87.71%           
  Complexity      634      634           
=========================================
  Files            71       71           
  Lines          2239     2239           
=========================================
  Hits           1964     1964           
  Misses          275      275           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.