Remove the per-element staging buffer in the BRNG fallback

[vlad-perevezentsev]

This PR proposes two changes to irk_uniform_bits_vec, neither of which alters generated values.

• The viRngUniform fallback taken by WH, MCG31, R250 and MRG32K3A assembles 16-bit halves in a 4096-entry stack tile instead of an mkl_malloc buffer of 8 or 16 bytes per element. That allocation was also unchecked in release builds, where assert compiles out, so a large request could crash instead of raising.
• The two overloads become one template on the word type, so the MKL_INT / 2 count cap of viRngUniformBits64 no longer lives in a copy of its own, far from the 32-bit limit it is contrasted with

For randint(0, 2**64, size=2**30 + 4096) on WH, peak memory drops from 24.05 to 8.05 GiB and time from 12.6 to 5.6 s, matching the native path in memory. Generation is also 1.6–3x faster for 4M–16M elements on all four BRNGs, and unchanged below that.