Merge pull request #12162 from AlexFigas/feat/libc-aligned-array-operators

feat(libc): export the remaining aligned and array operator new and delete overloads

docs/dev/debt/libc.md

@@ -109,6 +109,8 @@
 - [strtof and wcstof](../../../core/libs/prx/libc/src/Strings.cpp) (libc) round to the nearest float, where the console's `_Stof` in `libSceLibcInternal.sprx` (0x734a0) does not always: about a third of decimal inputs within a relative 2^-51 of a point halfway between two floats, and some hexadecimal inputs with more than 24 significant bits, come out one unit apart in the last place (`1e-40` is `0x1.16c3p-133` there and `0x1.16c2p-133` here). The console's own rounding is not reproduced
 - [strtof and wcstof](../../../core/libs/prx/libc/src/Strings.cpp) (libc) - `ERANGE` follows the `libc.prx` bundled with PPSA08427 (J03547016) and PPSA03671 (J03848509), both run side by side with this libc on a Linux host: a decimal input whose result is a nonzero subnormal sets it only when inexact, a hexadecimal one only when it lies below the smallest subnormal and rounds up to it, and underflow to zero and overflow set it. Both builds also take a `NAN(...)` payload in `strtof` from the sequence's leading digits (`nan(0x123)` is `0x7fc00123`, `-nan(5)` is `0xffc00005`); the sign is kept here, the payload is not. `libSceLibcInternal`'s `_Stof` differs: no `ERANGE` for hexadecimal input, a positive `nan`, and after a minus sign a failed parse is `-0.0` for a lone sign and `+0.0` before an `inf`/`nan` spelling, the reverse of `libc.prx`. Decimal `strtod` sets `ERANGE` for a nonzero subnormal in J03547016 but not in J03848509 or `libSceLibcInternal`, whose result is kept.
 
+- [operator new](../../../core/libs/prx/libc/src/StdioExtras.cpp) (libc) - the `nothrow` forms return null only for `std::bad_alloc`, where the `libc.prx` bundled with PPSA21564 (build `J03093718`) catches every exception there, so a new handler that throws anything else ends the process instead of returning null. The aligned forms throw `std::bad_alloc` for an alignment of zero or one that is not a power of two, where that binary calls `abort`.
+
 - [Guest time conversions](../../../core/libs/prx/libc/src/Time.cpp) rewrite the host process `TZ` when a guest changes its own `TZ`, so other host consumers of that variable (including libSceRtc local time) shift too; code outside libc does not take the zone lock. Zone syntax follows the host C runtime, not FreeBSD: Windows only accepts the `std offset [dst]` form, treats an empty guest `TZ` as `UTC0` and reads `:Europe/Berlin` as an abbreviation (for example `rop`); only Linux honours zone file names. If the guest accepts a `TZ` value that the host rejects (on Windows, for example, one longer than the host environment limit), conversions throw `std::runtime_error`. A `tm_zone` saved before a `TZ` change can refer to storage now holding the new zone name; `strftime` does not read it.
 - [cbrtf](../../../core/libs/prx/libc/src/Math.cpp) (libc) is FreeBSD msun's [`s_cbrtf.c`](https://github.com/freebsd/freebsd-src/blob/release/11.4.0/lib/msun/src/s_cbrtf.c): a 5-bit first guess from the exponent bits, then two Newton steps in double precision. It sets no `errno`, as the `libc.prx` that titles ship does not. Floating-point exceptions are not compared. It gives the same bits as the `cbrtf` of the `libc.prx` bundled with PPSA07923, PPSA05916 and PPSA13934 for all 2^32 arguments, run on a Windows host.
 - [logf](../../../core/libs/prx/libc/src/Math.cpp) (libc) is FreeBSD msun's [`e_logf.c`](https://github.com/freebsd/freebsd-src/blob/release/11.4.0/lib/msun/src/e_logf.c), with `vzero` read at run time as msun's `volatile` read does. As in the `libc.prx` that titles ship, `errno` is set to `ERANGE` for ±0 and to `EDOM` for every argument with the sign bit set, negative NaNs and -inf included; positive NaNs and +inf leave it unchanged. Floating-point exceptions are not compared. It gives the same bits and `errno` as the `logf` of the `libc.prx` bundled with PPSA07923 and PPSA05916 for all 2^32 arguments, run on a Windows host. The FMA-contracted build bundled with PPSA13934 differs from it by 1 ulp on 961,881 arguments, so no single port matches every title's last bit.