perf: add opt-in precompiled headers for chain/protocol/wallet

[chiliec]

Adds precompiled-header (PCH) support for the three heaviest libraries, gated behind a new ENABLE_PCH option (default OFF). No source code touched.

Motivation

cmake_minimum_required is already 3.16, which has native target_precompile_headers, but nothing used it. Every translation unit in chain, protocol, and wallet re-parsed the same large template-heavy third-party headers from scratch. An include-frequency scan of each library's sources shows the cost concentrated in:

• chain — Boost.MultiIndex containers (multi_index_container, ordered_index, hashed_index, composite_key) + FC serialization/reflection
• protocol — FC reflection / static_variant + the operation type machinery
• wallet — Boost.MultiIndex, Boost string/range algorithms, FC api/reflection

What this does

• New libraries/{chain,protocol,wallet}/pch.hpp, each listing only stable third-party headers (Boost, FC, std). Project headers are deliberately excluded — a churning header in the PCH would force a full-library rebuild on every edit and defeat the purpose.
• target_precompile_headers(<lib> PRIVATE pch.hpp) in each library, wrapped in if(ENABLE_PCH).
• New root option(ENABLE_PCH ... OFF).

Safety

• Default OFF — cannot affect existing builds, Docker images, or CI until a maintainer opts in.
• ccache-aware: PCH and ccache conflict unless ccache is told not to hash the PCH. When ENABLE_PCH=ON, ccache is active, and CCACHE_SLOPPINESS lacks pch_defines,time_macros, CMake emits a warning (otherwise ccache silently disables its cache for PCH TUs — you'd lose the ccache win without gaining the PCH one).

Usage

cmake -S . -B build -DENABLE_PCH=ON
# if you use ccache:
export CCACHE_SLOPPINESS=pch_defines,time_macros

Validation

I don't have Boost dev libs or the submodules in my environment, so a full tree build wasn't possible. What I did verify with real cmake:

• The ENABLE_PCH logic executes correctly: OFF prints a silent status line; ON prints status and the ccache warning only when CCACHE_SLOPPINESS is unset; ON + sloppiness is clean.
• The target_precompile_headers wiring itself was proven end-to-end on a standalone CMake project mimicking the same shape — 24 TUs sharing one heavy header set. Single-threaded build: 16.84s → 5.73s (2.9×), with a real .pch artifact produced. (std headers stand in for Boost there; Boost.MultiIndex + FC reflection parse costs are at least as high, so this is a conservative proxy — not a measurement of this repo's actual numbers.)

Ask for the maintainer: run a real before/after on a build box —

# baseline
cmake -S . -B b_off && time cmake --build b_off -j$(nproc)
# with PCH
CCACHE_SLOPPINESS=pch_defines,time_macros cmake -S . -B b_on -DENABLE_PCH=ON && time cmake --build b_on -j$(nproc)

and confirm the win holds on this codebase before considering flipping the default.

[On1x]

Folded into the pm branch (our mainnet-bound line, PR #124) to keep the HF14 work consistent and avoid master↔pm divergence. Full CI Docker build on pm is green. Closing here since the change now lives in pm; it'll reach master when pm merges. Thanks!