Add PIT Timer, IDT Fixes, and Deferred Kernel Timing Output

[amorrisCS]

This PR introduces PIT-based timing support and integrates it with the kernel, along with fixes to IDT handling and timing correctness.

Features

• Adds PIT (Programmable Interval Timer) initialization at configurable frequency (~1000 Hz)
• Implements IRQ0 handler and tick counter
• Provides kernel_get_ticks_ms() for millisecond timing

Key Fixes
IDT segment selector fix:

• Removed hardcoded 0x08
• Now reads %cs at runtime in idt_init() to prevent GP faults on iret

Timer overflow fix:

• Resolved uint32_t overflow in kernel_get_ticks_ms() that caused wraparound after ~3.6 seconds

PIT divisor bug:

• Fixed shadowed variable so timing calculations use the correct divisor

Design Improvements

• Moved serial output out of IRQ context:
• IRQ handler sets a print_pending flag
• Printing handled in kernel_main
• Keeps interrupt handler fast and avoids timing drift
• Added safe uint32_t serial formatting via pre-buffered output

Other

• Cleaned up unused includes
• Rebased onto main to resolve merge conflicts

Result

• Stable interrupt returns (no GP faults)
• Accurate and consistent time tracking
• Safe, minimal interrupt handler design
• Clean base for future scheduling/timing features

[LordHerdier]

This looks really good! Thank you for all of your work on the interrupt handling!

I'm approving this PR to merge with a couple of notes about bugs that will need to be fixed in the future:

Bugs

isr.s:12 : default_stub is unsafe for exception vectors that push an error code

default_stub just does a bare iret, but CPU exception vectors 8, 10–14, 17, 21, 29, and 30 push an error code onto the stack before entering the handler. A bare iret from those vectors pops the error code as the return %eip, misaligns the stack, and will likely triple-fault. The stub is currently installed on all 256 IDT entries, so any one of those exceptions firing (e.g. a page fault during early init) will cause a hard crash instead of a clean fault. Should at minimum have a separate error_stub that pops the error code before iret, and install it on the appropriate vectors.

Minor Issues

1. idt_set_gate has a hidden dependency on idt_init() having been called
   idt_set_gate uses kernel_cs (idt.c:48) which is only populated inside idt_init(). Calling idt_set_gate before
   idt_init() silently installs gates with selector 0, which would GP-fault on any interrupt. The dependency isn't
   documented and isn't enforced. At minimum a comment on idt_set_gate saying "must be called after idt_init()" would
   help. A future refactor could fold gate setup into idt_init() where the call ordering is explicit.

2. build.sh special-cases isr.s rather than handling all .s files generically
   The existing loop assembles .c files; there's a separate if [ -f src/isr.s ] block for assembly. If more .s files are
   added later, each would need another special case. A glob over src/.s (matching how src/*.c is handled) would be
   more maintainable.

3. C-style () vs (void) in new function declarations
   void idt_init(), void pic_remap(), etc. — in C (not C++), empty parens mean "unspecified parameters", not "no
   parameters". All the new functions that take no arguments should use (void). This is consistent with what the existing
   codebase does (e.g. serial_init(void)).

4. Comment style inconsistency
   The existing codebase uses /* */ block comments; new code in kernel.c uses // line comments. Minor, but worth keeping
   consistent.

---

I'm going to create a couple Jira tasks to tackle these that we can both work on. For now though, this implementation is good!