Some printers (confirmed: NT-5890K) silently reset their active code page
back to the factory default after a font switch (`ESC M n`). `MagicEncode`
was unaware of this hardware-side reset: its cached `self.encoding` remained
stale, causing subsequent `text()` calls to skip `CODEPAGE_CHANGE`. Non-ASCII
characters were then sent in the previously-active encoding (e.g. CP1257)
but interpreted by the printer under its default code page, producing
garbled output.

`ESC @` (`hw("INIT")`) is defined in the ESC/POS spec as a full printer reset
that restores all settings to factory defaults, including the active code
page — so the same fix applies there by spec.

Fix: add `MagicEncode.reset_encoding()` which clears both `self.encoding`
and `self.encoder.used_encodings`, and call it from `set()` after every
font change and from `hw()` after `INIT`. This forces a fresh code page
selection and a `CODEPAGE_CHANGE` re-emission before the next text output.

Why `used_encodings` must also be cleared:
`self.encoding = None` is enough to ensure a `CODEPAGE_CHANGE` is emitted.
However, `used_encodings` biases `find_suitable_encoding()` toward previously-
used code pages. After a reset, that preference is stale: on NT-5890K the
previously-used CP1257 does not function correctly after `ESC M`, so
`MagicEncode` would re-select it, emit `CODEPAGE_CHANGE` → CP1257, and send
e.g. `ü` as `0xFC` — wrong in the printer's default code page. Clearing
`used_encodings` removes the stale bias and lets slot-number ordering take
over, landing on CP850 where `ü` = `0x81`, a byte that is correct in
virtually every Western code page regardless of whether the printer
honours the code page switch.

Follow-up: `used_encodings` could be removed from `MagicEncode` entirely.
`self.encoding` is the mechanism that actually avoids redundant switches: it
keeps the current code page as long as it can encode the next character,
without consulting `used_encodings` at all. `used_encodings` is only consulted
when a switch is already unavoidable — at which point it cannot prevent
any switch, it can only gamble on which encoding might be needed again
later. That saves at most one future switch in the rare case where the
same non-default code page is needed again after having been forced away.
Removing `used_encodings` would also make `reset_encoding()` unnecessary:
with only `self.encoding` to clear, callers would just write
`self.magic.encoding = None` directly, with no need for a helper method.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…cstring