feat: unified generation options, xs:pattern/xsi:type semantics, base-field inlining fixes, and phf tag dispatch

[nth-bailey]

Summary (10 commits)

CLI / schema semantics

• a5337af feat(cli): unify generation options and add target-aware completion
• 05466ba feat(schema): preserve xs:pattern OR/AND semantics and enforce patterns in all 7 codegens ([Schema Parser & Codegen] xs:pattern OR-within-restriction semantics + missing pattern enforcement in Rust/Go/C++ codegen #54)
• f82abd7 refactor(cli): remove deprecated legacy flags and manifest keys
• 2c6e611 feat(schema): xsi:type polymorphic dispatch for abstract complexTypes ([Schema Parser & Codegen] xsi:type polymorphic dispatch for abstract complexTypes #53)

Codegen correctness fixes

• 93713b3 fix(rust-codegen): inline xsd:extension base fields in generated structs
• 5e559ee fix(java-codegen): inline base fields in plain record mode
• c16e57d fix: simpleContent text codecs, split-text accumulation, and transcoder entity refs

#49: perfect-hash element tag dispatch

• dce03c1 feat(rust-codegen): --feature phf emits a per-struct __{Struct}ElementId enum + phf::Map dispatch at Event::Start/Event::Empty; default output stays byte-identical. Includes benches/tag_dispatch.rs (criterion, tiers 16/120/600/1500, hit+miss), scripts/perf_stat.{c,sh} hardware counters, tests (18/18 rust codegen incl. 3 new phf tests, 35/35 CLI), verify_codegen.sh phf step, and full results/tradeoff analysis in docs/benchmarks/rust-phf-dispatch.md.
  • Measured: throughput crossover ≈ 835 elements (phf wins above; miss crossover ≈ 400), match's L1I misses grow 277× from 120→1500 tags while phf stays flat; recommendation documented: auto-enable at ≥ ~900 elements.
  • Small-schema cost: +0.04 % .text, +144 B .rodata, rebuild delta within noise.
  • Compile-time/size at the 600/1500 tiers exceeded the safe RAM cap for every variant (scales with field count, not dispatch) → deferred to [Benchmark] Large-schema compile-time & binary-size comparison (600/1500 elements) deferred from #49 #55.
• 020794f docs: link that caveat to [Benchmark] Large-schema compile-time & binary-size comparison (600/1500 elements) deferred from #49 #55; fix garbled memcap sentence in SKILL.md

Review fixes

• d6f567d enforces full-value XSD patterns in all seven generators, resolves xsi:type values against in-scope namespace bindings, rejects invalid boolean simpleContent, refreshes Python subclass dispatch after schema caching, and adds regression coverage and documentation.

Benchmark host safety

• scripts/memcap.sh: portable 60 %-of-available-RAM guard (systemd scope + ulimit -v fallback, nesting-aware) wired into benchmarks/run_all.sh, benchmarks/cli/benchmark.sh, scripts/perf_stat.sh so heavy builds OOM-kill in-cgroup instead of freezing 8-GiB dev boxes. CI never references those entry points.

Verification (full AGENTS checklist, all green)

• cargo fmt --check · cargo clippy --workspace --all-targets -- -D warnings (0 warnings)
• cargo test --workspace (all suites)
• ./scripts/test_codegen.sh (all 7 language suites + CLI integration)
• ./scripts/verify_codegen.sh (7 ecosystems × backends smoke, incl. new --feature phf + ELEMENT_DISPATCH assertion)
• Python gate: maturin develop · ruff check + ruff format --check · pytest --cov=polyxml --cov-branch --cov-fail-under=100 → 100.00 %, 72 passed

Issues

• Closes [Codegen: Rust] Compile-Time Perfect Hashing (phf) and SIMD Token Dispatch #49: opt-in phf codegen, dispatch benchmarks, hardware counters, and documented threshold recommendation. [Benchmark] Large-schema compile-time & binary-size comparison (600/1500 elements) deferred from #49 #55 tracks large-schema compile time and binary size; [Benchmark] Validate phf dispatch threshold with end-to-end XML decoding #57 will test that threshold in end-to-end XML decoding before any automatic selection.
• Closes [Schema Parser & Codegen] Audit: 10 edge-case XSD handling items from competing parser issue trackers #51: all ten audit items are resolved or covered by the linked feature work.
• Closes [Schema Parser & Codegen] xsi:type polymorphic dispatch for abstract complexTypes #53: native xsi:type dispatch and serialization are covered, including the review fixes above.
• Closes [Schema Parser & Codegen] xs:pattern OR-within-restriction semantics + missing pattern enforcement in Rust/Go/C++ codegen #54: grouped XSD pattern semantics and enforcement across all seven generators are covered.
• [CLI & Architecture] Consolidate target-specific boolean flags into unified options (--backend, --style, --feature) #50 remains open: unified options are implemented, but its requested legacy migration aliases and warnings are unresolved. [CLI] Decide compatibility policy for unified options #58 tracks the compatibility decision and implementation. Python plain class, C# mutable struct, and cold-start measurement remain in [CLI & Architecture] Consolidate target-specific boolean flags into unified options (--backend, --style, --feature) #50.
• [Target: Wasm & TS] WebAssembly Streaming Runtime (polyxml-wasm) for Node.js, Bun, and Browsers #47 (polyxml-wasm) and [Codegen: Python] Ahead-of-Time (AOT) Compiled PyO3 Schema Extension (--aot) #48 (AOT --aot PyO3) remain open under their own issues.