fix(pic): unmask cascade IRQ2 when enabling a slave IRQ

[kudasaixc]

pic_remap() masks every IRQ line (0xFF) including IRQ2, the cascade input for the slave PIC, and nothing ever unmasks it. Any IRQ in the 8-15 range (RTC, PS/2 mouse, ATA, ...) could therefore never be delivered, even after calling pic_unmask_irq() for it. This unmasks the cascade line on the master whenever a slave IRQ is enabled.

[kudasaixc]

PoC

pic_remap() ends by masking everything:

outb(PIC1_DATA, 0xFF);
outb(PIC2_DATA, 0xFF);

That includes IRQ2 on the master : the line the slave PIC is cascaded onto : and nothing ever clears it. So no slave interrupt (IRQ 8–15) can reach the CPU, even after you "enable" it:

// try to enable the RTC (IRQ 8)
irq_register_handler(8, rtc_callback);
pic_unmask_irq(8);           // clears IRQ8 in the SLAVE mask only
// rtc_callback never fires: IRQ2 on the master is still masked,
// so the slave's output is blocked before it reaches the CPU

pic_unmask_irq(8) computes port = PIC2_DATA and clears bit 0 of the slave mask, but the master never learns it should listen on IRQ2. Any future driver on IRQ 8–15 (RTC, PS/2 mouse @ IRQ12, ATA @ IRQ14/15) is silently dead. Fix clears IRQ2 on the master whenever a slave line is unmasked.

[im-nymii]

LGTM