[inskr]

Fixes #7576

Summary of changes:

• stm32_qspi_nor_program() and stm32_xspi_nor_program() now stop at the page boundary instead of handing the whole length to one page program command.
• The 256-byte page constant sits next to each driver.

I put the fix in the shared driver instead of in each board's extflash_program(): nine extflash_driver.cpp copies call into these two functions (the issue lists five), and the lower layer is where the page size is known.

The per-board loops still slice by length. That is harmless now, the lower layer re-slices on the page boundary.

How it was checked

Host harness: the patched function body compiled against a fake HAL that wraps like real NOR flash. HAL_QSPI_Command() / HAL_QSPI_Transmit() are stubbed, and every page program command is checked for offset + len > 256.

Current master:

addr=0x100 len=300 -> cmds=1 cross_page=1 CORRUPT (44 bytes wrong, first +256)
addr=0x180 len=300 -> cmds=1 cross_page=1 CORRUPT (171 bytes wrong, first +128)
addr=0x1F0 len=32  -> cmds=1 cross_page=1 CORRUPT (16 bytes wrong, first +16)

With this patch:

addr=0x100 len=300 -> cmds=2 cross_page=0 ok
addr=0x180 len=300 -> cmds=2 cross_page=0 ok
addr=0x1F0 len=32  -> cmds=2 cross_page=0 ok

Both changed functions are STM32 HAL only, so this is as far as it goes without a radio. The normal update path writes page aligned from a sector start, so a hardware test would not show a difference either way.