[PHP] fix: restore boxed process construction helpers

[brandonpayton]

Why

A Process record is large, and constructing one as a value in a caller's stack frame places the whole struct on that frame. On wasm the call stack is small and non-growable, so building a Process inline on deep paths (fork/exec/spawn) risks stack overflow. A heap-allocating constructor keeps large process records off the caller's frame. The PHPT harness and PHP subprocess tests drive exactly those deep process-creation paths, so this is a platform robustness fix, not PHP-specific code.

A companion Default for WasmStat gives the kernel an ergonomic, consistent way to build stat records (used by procfs and other synthetic-stat sites), so callers can spread-construct instead of duplicating every field.

What

• crates/kernel/src/process.rs: add Process::new_boxed() (marked #[inline(never)], returns Box<Self> from a private new_value) so call sites can heap-allocate the record; Process::new() now delegates to it.
• crates/shared/src/lib.rs: add impl Default for WasmStat (all-zero fields) enabling ..WasmStat::default() construction.

Provenance

Supersedes fork-headed PR #727; identical head commit 504c802110c0c623152d95b22195e0873f48075d, re-pointed onto an Automattic/kandelo branch. No code changed from #727.

[brandonpayton]

Closing #747 after Batch-2 review as superseded/incomplete; it will not be replayed into #871.

Review findings:

• The old commit conflicts with the current explicit-stdio constructor (Process::new_with_stdio).
• Process::new_boxed has no production caller. Process::new immediately dereferences the box and the process table still stores Process by value, so the patch adds allocation/move/deallocation without preserving boxed ownership.
• The earlier complete stack-avoidance design used Box::new_uninit() with direct field initialization and boxed process-table values across create/fork/exec/spawn. This split restores none of that end-to-end ownership.
• The added WasmStat::Default is unrelated, unused, and omitted from the title. Batch 1's procfs repair deliberately removed the redundant default construction that once motivated it.
• No current Batch-2 constituent depends on either helper, and there is no focused Wasm stack reproducer or measurement supporting the PR's robustness claim.

The original head does compile, and a throwaway mechanical adaptation to the current constructor also compiled and passed the structural ABI check; the rejection is therefore about correctness, scope, and evidence—not an unresolvable merge conflict.

If kernel Wasm stack depth is reproduced as a current defect, it should get a fresh focused PR that covers every constructor including explicit stdio, retains boxed ownership through the process table and fork/exec/spawn paths, omits unrelated stat convenience APIs, and validates the actual failure on Node and Chromium.