feat(builtins): buf_resample_linear — endpoint-inclusive linear resample kernel (#603)

[InauguralPhysicist]

Closes #603

What

Third leg of the DAW audio-I/O train (after #597 mix kernels and #602 PCM codecs): DeslanStudio's WAV import spent 3.9 s resampling a 50 s stereo file on the N3350 in ab_resample_linear's interpreted per-sample loop.

buf_resample_linear of [src, dst_len] → NEW buffer with exactly the consumer's endpoint-inclusive mapping:

pos = i * (n - 1) / (dst_len - 1)    (0 when dst_len == 1)
lo = floor(pos); hi = min(lo + 1, n - 1); frac = pos - lo
out[i] = src[lo] * (1 - frac) + src[hi] * frac

num_guard per step in VM evaluation order — bit-identical to the interpreted loop, pinned by the differential leg in tests/test_buf_resample.eigs (upsample/downsample/identity/dst 1/n 1 shapes on seeded pseudo-random data).

Doc note in BUILTINS.md: the kernel is linear interpolation, not Fourier/sinc — the consumer-documented divergence from scipy.signal.resample (no anti-aliasing).

Edges: dst_len 0 → empty buffer; negative/non-numeric dst_len raises value; empty src with dst_len > 0 raises value; dst_len > INT_MAX raises limit. Allocation charges the #292 sandbox budget; sandbox pure-compute allowlist; freestanding-safe.

Perf (n=5 medians, N3350, 1,102,500 → 1,200,000 samples, pinned to one core)

leg	per pass
interpreted (JIT)	1048 ms
interpreted (JIT off)	1986 ms
buf_resample_linear	60.3 ms (~17x / ~33x)

Gates

• Release suite 2890/2890, 0 failed (includes the new 5-check section).
• ASan/UBSan detect_leaks=1: 2893/2894, leak tally 0 — single failure is the known invariant_weak [97] 60 s wall-guard marginality under ASan on this box (71 s in isolation; unrelated).
• Doc gates [99]/[99b] green.

🤖 Generated with Claude Code