refactor(ofdft): reduce solver global state dependencies

[zzlinpku]

Summary

Reduce ESolver_OF's direct dependence on global runtime state while preserving the existing initialization and numerical behavior.

• Snapshot the six PARAM.globalv settings used by OFDFT into OFDFTRuntimeConfig at the solver initialization boundary, then use that configuration for charge initialization, output paths, force, and stress calculations.
• Pass the warning stream explicitly through the OFDFT convergence, direction-check, and line-search methods. The base OFDFT and OFDFT-TDDFT runners provide the stream at their orchestration boundary.
• Keep initialization stage markers and XC static-state access unchanged. XC lifecycle and mutation behavior remain a separate follow-up evaluation.

Validation

• cmake --build build --target abacus_basic_para -j2 — passed.
• ctest --test-dir build -R '^MODULE_PW_OfDFT$' --output-on-failure — passed (1/1).
• ctest --test-dir build -R '^07_OFDFT$' --output-on-failure — 169 of 173 cases passed. Four cases require optional build support unavailable in this configuration: 04_OF_KE_CPN5 and 06_OF_KE_MPN require ENABLE_MLALGO; 22_OF_LibxcPBE and 26_OF_MD_LibxcPBE require LibXC support.
• git diff --check — passed.

[mohanchen]

@ABACUS-PR-BOT could you review the PR?

[ABACUS-PR-BOT]

Static review of 4193ee4: I found no correctness issue in the snapshot and warning-stream changes, including the OF-TDDFT callers. Two non-blocking follow-ups are below. Whitespace and trusted governance checks completed without errors; no builds or runtime tests were run, and CI diagnostics were unavailable.

[ABACUS-PR-BOT]

Non-blocking: Please use <iosfwd> here, since this header only declares std::ostream& parameters and does not need the complete stream definition. The output operations are implemented in the .cpp files. This can be handled in a follow-up.

[ABACUS-PR-BOT]

Non-blocking: Could you add a focused regression test for the captured settings and an injected std::ostringstream warning sink? MODULE_PW_OfDFT only compiles KEDF code, and the calculation cases do not distinguish the supplied sink from GlobalV::ofs_warning or a retained snapshot from later global reads. The existing integration coverage is useful; this additional coverage can follow after merge.

[mohanchen]

This PR may not be the best solution, I will close it.