Offer the letter keyboards in alphabetical order

[oroderico]

Closes #9.

Summary

QWERTY is a poor default for someone who does not touch-type: on a ring of keys walked with two buttons, a scattered alphabet means hunting for every letter. Settings now chooses, beside Flip Orientation and Brightness — Keyboard: QWERTY or Keyboard: ABC.

The layouts were already strings, and seedtool_app.c already said why: every key is found by its character rather than by where it sits. Nothing depends on position — the cursor walks a flat ring, the renderer centres each row from the string itself, and layout_key_index searches by character. The alphabetical word layout is 10/9/8 like the QWERTY one, so it fits the three rows the keyboard is drawn in without the renderer knowing them apart.

Covers the word keyboard and the passphrase letter pages. The number keyboard has no alphabetical form. The passphrase digit and symbol pages are shared between the two arrays rather than duplicated — punctuation has no alphabetical order, and a second copy would be a second place for a character to go missing.

It cannot change mid-session

Settings is reachable only from the Origo menu, and the wallet has no row that returns there, so no session can have half a passphrase typed on one arrangement and half on the other. The presses would spell the same string anyway — keys are found by character — but the menu tree makes that structural rather than something the reader has to be careful about.

The checks matter more than the feature

A layout typed out by hand is exactly where a missing letter comes from, and a missing letter makes some BIP39 words impossible to enter: the reader reaches for a key that is not there and has no way to finish a restore.

• layouts_are_complete now takes a layout rather than naming one, and runs over both word arrangements and both passphrase arrays. Each array is checked for full printable-ASCII coverage on its own, not pooled — a character present in only one arrangement is as unusable as a missing one for whoever picked the other.
• The Python source-parse test does the same from the source text, and a new test pins both word layouts to exactly the 26 letters.
• Both were proved by deleting the s from the alphabetical layouts and confirming the self-test and two Python tests fail, naming the letter.

Size

+336 bytes, taking the image to 99.6% of the ceiling with 1184 free. More than the ~63 bytes the strings alone suggested — the rest is the second passphrase array, the two accessors and the extra Settings row. Flagging it rather than burying it: the ceiling is close enough that the next change of any size needs the budget looked at first.

Verified

• Host self-test, 50 Python tests, sanitized build.
• Firmware build against the pinned IDF 5.5.4, ELF audit passing at 300896 bytes.
• Hardware: flashed and walked after merge.

[oroderico]

Reverted in 99c2dad — not wanted yet. Diff stays here if it is picked up later.