feat: add musl libc runtime and native C/Python workflows

[TRSWNCA]

Summary

• Add a patched musl libc runtime that forwards Linux syscalls into AlloyStack LibOS hostcalls.
• Add a C AsBuffer API for reference-based workflow data exchange, including non-destructive buffer borrowing.
• Port wordcount, parallel sort, and long-chain C workloads from Wasmtime to native musl DSOs.
• Build CPython 3.11.9 as a static musl library from the pinned third_party/cpython submodule.
• Add a native built-in pyas module and pass workflow arguments through Python sys.argv.
• Add native musl CPython workflows and Wasmtime versus musl end-to-end comparison recipes.
• Add directory handles, getdents64, path stat, clock, random, futex, mmap, and related libc compatibility support.
• Serialize FATFS metadata access so concurrent native CPython instances do not violate fatfs internal borrowing rules.

Native Python workflows

The following native musl configurations were added:

• wordcount C1 and C3
• parallel sort C1 and C3
• function chain N5

Concurrent CPython 3.11 instances currently use distinct DSO copies so each workload receives an independent CPython global runtime.

Validation

• Built the musl runtime and CPython with just musl_cpython.
• Ran native musl CPython wordcount C1 and compared reducer output with an independent recount.
• Ran native musl CPython parallel sort C1 and its ordering checker.
• Ran native musl CPython function chain N5.
• Ran native musl CPython parallel sort C3 with three concurrent CPython DSOs.
• Re-ran the existing Wasmtime CPython wordcount, parallel sort, and function-chain workflows after restoring the legacy readdir hostcall name.
• Confirmed the native CPython DSO has no glibc DT_NEEDED dependency.

Experiment recipes

• just musl_c_end_to_end_latency
• just musl_py_end_to_end_latency
• just py_end_to_end_latency_compare

The existing Python latency recipe now preserves and prints the complete asvisor error instead of hiding it behind grep.

Current limitations

• Py_FinalizeEx and post-main allocator teardown are not yet safe. CPython is treated as an isolation-lifetime runtime and cleanup is deferred to isolation unloading.
• The compatibility pyas.access_buffer API copies into an existing Python buffer. pyas.borrow_buffer provides a zero-copy read-only memoryview for workloads migrated to the new API.
• CPython C3 concurrency currently uses separate DSO copies rather than subinterpreters.
• wordcount C3 requires an input dataset compatible with the fixed 25,677-byte script buffer. The current local image contains approximately 35 MB fake_data_1 and fake_data_2 files, whose reduced output exceeds that capacity.
• Full performance recipes were not run as part of this implementation.

[TRSWNCA]

不需要borrow_buffer的语义

[TRSWNCA]

不需要borrow_buffer的一套语义，同时musl python的接口还是在采用copy的方式，没有零拷贝

[TRSWNCA]

Benchmark

Tested on:

• CPU: Intel(R) Core(TM) i7-10700 (16) @ 4.80 GHz
• Disk: ST2000DM001-1ER164 (SSD) 1.8TB

Word Count

Unit: ms

runtime	parallelism	10MB	100MB	300MB
rust	C1	79.104	757.638	2268.314
rust	C3	36.503	314.195	922.197
rust	C5	31.699	227.090	663.979
wasm-c	C1	299.279	2387.993	7095.795
wasm-c	C3	200.901	890.402	2467.586
wasm-c	C5	274.950	751.190	1632.348
wasm-py	C1	1092.912	8955.991	26754.322
wasm-py	C3	869.074	3572.548	9410.658
wasm-py	C5	1132.036	3016.441	6964.269
musl-c	C1	147.171	2248.459	13052.668
musl-c	C3	75.075	781.759	3053.358
musl-c	C5	51.720	523.537	1759.068
musl-py	C1	493.236	4339.102	14037.764
musl-py	C3	352.915	1583.907	4461.630
musl-py	C5	419.338	1283.076	3081.894

Parallel Sort

Unit: ms

runtime	parallelism	1MB	25MB	50MB
rust	C1	13.350	328.367	700.638
rust	C3	10.838	199.952	453.829
rust	C5	10.402	188.218	364.902
wasm-c	C1	591.099	16281.043	N/A
wasm-c	C3	220.988	5980.738	N/A
wasm-c	C5	151.141	4114.035	N/A
wasm-py	C1	584.279	10230.233	20992.677
wasm-py	C3	793.674	5649.050	10980.069
wasm-py	C5	1149.108	5292.692	9571.251
musl-c	C1	175.673	4945.973	10471.834
musl-c	C3	62.404	1662.802	3473.104
musl-c	C5	41.252	1072.429	2241.401
musl-py	C1	312.996	6346.976	12864.340
musl-py	C3	352.578	3702.550	7453.610
musl-py	C5	458.735	3283.261	6202.339

Long Chain

Unit: ms

runtime	chain_len	1MB	64MB	256MB
rust	n5	1.360	26.047	114.898
rust	n10	1.984	48.788	216.746
rust	n15	2.797	70.335	324.416
wasm-c	n5	5.202	34.414	116.353
wasm-c	n10	8.408	76.533	449.604
wasm-c	n15	12.116	117.584	823.776
wasm-py	n5	205.243	288.193	533.810
wasm-py	n10	207.786	391.917	978.220
wasm-py	n15	207.361	496.095	1388.698
musl-c	n5	1.537	1.326	1.328
musl-c	n10	1.862	1.678	1.583
musl-c	n15	2.306	1.852	2.368
musl-py	n5	70.525	74.743	87.599
musl-py	n10	70.747	74.415	88.712
musl-py	n15	70.564	75.283	88.309