Defer RISC-V callee-saved registers on no-switch timer interrupts

[zhouzhou-code]

Description

The dedicated RISC-V machine-timer handler currently saves and restores the
complete integer context even when xTaskIncrementTick() returns pdFALSE.
The called C code must preserve s0-s11 according to the RISC-V ABI, so those
registers do not need to be written to the task stack when execution returns to
the interrupted task.

This change keeps the existing task-frame layout and optional-context behavior,
but splits integer context handling into caller-saved and callee-saved parts:

1. Save caller-saved registers on machine-timer entry.
2. If no context switch is required, restore only those registers and return.
3. If a context switch is required, save s0-s11 into their reserved frame
   slots before calling vTaskSwitchContext().

FPU, VPU, and chip-specific additional contexts retain their existing save and
restore semantics. Other interrupt, exception, unified-trap, and yield paths
continue to save the complete context.

For RV64, the no-switch path avoids 12 stores and 12 loads, reducing stack
traffic by 192 bytes per tick.

Test Steps

A standalone QEMU virt M-mode test uses different sentinel values in
s0-s11 for two equal-priority tasks and checks them after 2,000 timer ticks.
The following configurations passed against the current main branch:

• no switch
• task switch on every tick
• FPU, with and without task switching
• VPU, with and without task switching
• synthetic chip-specific additional context
• FPU + VPU + additional context + task switching

The modified portASM.S was also assembled successfully for RV32 (ilp32)
and RV64 (lp64d). Disassembly confirms that the pdFALSE path contains no
s0-s11 stores or loads and that the pdTRUE path stores all twelve registers
before vTaskSwitchContext().

The same implementation was previously validated on a JH7110/U74 in M-mode:

• baseline and optimized single-task and two-task tests passed 1,000,000 ticks
• 10,000 no-switch entry samples: median 59 -> 50 cycles; mean 155.281 ->
  144.746 cycles
• P99 and maximum did not improve

Checklist:

• I have tested my changes. No regression in existing tests.
• I have modified and/or added unit-tests to cover the code changes in this Pull Request. (The standalone architecture test is maintained outside this kernel-only repository.)

Related Issue

Closes #1464

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud