Intercept cargo's -C strip in the rustc wrapper instead of predicting it

[nicoburns]

Summary

Fixes #5119 (wasm-opt SIGABRT "compile unit size was incorrect" on dx build --web --release, and more generally ~1.3MB of leftover std DWARF in release wasm).

Since #4966, dx injects --config profile.<x>.strip=false so manganis asset symbols survive the build, then strips manually in post_process_executable() based on get_strip_setting(). That prediction only honored an explicit strip key in Cargo.toml, missing Cargo's implicit default (since Rust 1.77, debuginfo is stripped when the profile's debuginfo is off — as release is by default). The un-stripped DWARF then gets rewritten by wasm-bindgen and crashes binaryen's DWARF re-emitter.

Instead of predicting what Cargo would do, this PR lets Cargo compute the strip setting and intercepts it in the rustc wrapper (RUSTC_WORKSPACE_WRAPPER), which dx already uses for Base/Fat builds:

run_rustc():
  strip = extract_strip_arg(&mut captured_args)   // removes -C strip=X / -Cstrip=X
  RustcArgs { args: filtered, envs, cwd, strip }  // new #[serde(default)] strip field
  // rustc is invoked without the flag, so symbols survive for asset extraction

post_process_executable():
  match recorded_strip_setting(&artifacts.mode) { // reads {tip}.bin.json from the scope dir
    "debuginfo" => rust-objcopy --strip-debug,
    "symbols"   => rust-objcopy --strip-all,
    _           => no-op,
  }

• profile_args() no longer injects strip=false, and get_strip_setting() is deleted.
• Since Cargo itself resolves the value, every source it considers is honored with no duplicated logic: profile inherits chains, .cargo/config.toml [profile] tables, CARGO_PROFILE_* env vars, and the implicit 1.77 default.
• Thin builds replay the captured args, which already have the flag removed, so patches stay un-stripped (post-processing only runs for Base/Fat). wasm_split builds are still never stripped.
• Bumped the rustc-wrapper scope version to 2 so stale captures (with -C strip still in args and no strip field) aren't replayed.

This is the more robust alternative to #5736, which instead mirrors Cargo's implicit default inside get_strip_setting() (and cannot see .cargo/config.toml / env-var overrides).

Verification

With a fresh cargo new app depending on dioxus (feature web), using dx built from this branch:

• dx build --web --release: output wasm has no .debug_* custom sections (on main, ~1.1MB of .debug_* remains).
• dx build --web (dev): debug sections retained.
• Custom profile with strip = "symbols": binary is fully stripped (--strip-all, no name section), assets still extracted.
• Custom profile with debug = true and no strip key: not stripped.
• CARGO_PROFILE_RELEASE_STRIP=none dx build --web --release: env override honored, debuginfo retained (the key case cli: mirror Cargo's implicit strip default when manually stripping binaries (fixes wasm-opt SIGABRT) #5736 can't handle).
• dx build --web --fat-binary: fat build path still works.

cargo fmt and cargo clippy -p dioxus-cli are clean.

Link to Devin session: https://dioxus.staging.devinenterprise.com/sessions/51b6fb74d9a0452985a2f6840ff2d2bd
Requested by: @nicoburns

[staging-devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring