feat(fspy-ipc-str): convert IpcStr and platform C strings directly

[wan9chi]

Motivation

The shm id still crossed the process boundary through OsStr: on Windows, IpcStr's UTF-16 bytes were decoded to WTF-8 (OsString) and re-encoded back to UTF-16 — two global-allocator allocations while the preload attaches under the loader lock. The preload is going std-free bottom-to-top, and the allocator is the bottom, so this lands the cross-platform arena together with its first consumer.

Built in commit order:

• fspy_nostd: VirtualAlloc/VirtualFree wrappers — kernel calls that never touch the CRT heap lock, usable under the loader lock.
• fspy_nostd_alloc: VirtualAllocator, the stateless Windows twin of MmapAllocator; the chunk pool and arena() now compile on every platform over a platform-selected page allocator.
• fspy_ipc_str: from_os_c_str borrows an OsCStr's units zero-copy, and to_os_c_string_in decodes into a caller-chosen allocator, reading Windows code units bytewise so IpcStr's unaligned representation never matters. Neither direction touches OsStr.
• fspy_shared: channel() serializes the id straight from the keeper's C string, and sender() decodes it with the arena — no OsStr round trip and no global-allocator use on the attach path.

---

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}

[github-actions]

fspy benchmark

linux

dynamic/launch             change  +1.78%  [ -4.05% ..  +7.79%]  overhead   +52.89%
dynamic/access             change  -0.41%  [ -7.52% .. +21.74%]  overhead    +5.90%
dynamic/access-relative    change  -0.74%  [-15.43% ..  +2.43%]  overhead   +50.67%
static/launch              change  +0.77%  [ -5.50% ..  +7.67%]  overhead  +154.05%
static/access              change  +0.42%  [ -3.21% ..  +4.99%]  overhead  +739.36%
static/access-relative     change  -0.01%  [ -1.88% ..  +1.27%]  overhead +1179.16%

macos

dynamic/launch             change  -0.12%  [ -4.66% ..  +3.90%]  overhead  +219.31%
dynamic/access             change  +0.35%  [ -4.70% ..  +6.55%]  overhead    +6.01%
dynamic/access-relative    change  +0.53%  [ -1.88% ..  +3.96%]  overhead  +259.77%

windows

dynamic/launch             change  -0.47%  [ -4.29% ..  +2.26%]  overhead   +27.38%
dynamic/access             change  +0.18%  [ -1.28% ..  +1.44%]  overhead    +1.51%
dynamic/access-relative    change  +0.55%  [ -1.08% ..  +2.16%]  overhead    +1.82%