docs(en): add 5 new translated docs (CVM/physical bench, rib-fix, runtime-fix)

Translate the 5 zh_cn-only documents into the English top-level

directory using the T2 strategy (structural alignment + key content

preserved verbatim: perf flame-graph root causes, diff snippets,

function signatures, 9 TC verdicts, NFR-1 matrix, KL table, DP

decision points, 6-commit chain). Privacy mask reuses x.x.x.39 etc.

New files (1,518 lines, ~92 KB):

  rib-fix-plan.md (101 lines)

    Plan for runtime-fix Phase 2: ff_ifconfig 'inet 192.168.1.1' fix.

    Carries the errno 55 = ENOBUFS correction (NOT EOPNOTSUPP, the

    Phase 1 mis-mapping); 4-way cross-verified rt_alloc signature

    table; 6-phase pipeline; DP-RT-FIX-1/2/3 (A/B/A default).

  cvm-bench-methodology.md (201 lines)

    Two-machine CVM benchmark methodology (server F-Stack data plane

    + control plane via ssh; client kernel/F-Stack pick-one). 10-step

    standard procedure (build both trees / app dual install / nohup

    --proc-type=primary / smoke / T1-T2-T3 / sed-based IP masking on

    AI-AGENT side / kill_process.sh / rtemap cleanup / tree switch /

    config restore + English commit message no-push). Cross-app

    verdict (helloworld / nginx / redis on lcore=4).

  physical-machine-bench-report.md (257 lines)

    Bare-metal baseline (Intel Xeon 8255C @ 2.5 GHz + Mellanox CX-5

    100 G + TencentOS 4.4 + Linux 6.6.98). External team data via

    iWiki 4021545579, distilled in-project. Headlines: helloworld

    single-core long-conn +10.24% (Req/sec 958,109 -> 1,056,178);

    nginx_fstack long-conn 1/2/4 cores +5.06% / +4.76% / +4.83%;

    nginx_fstack short-conn 1/2/4 cores -2.25% / -3.65% / -6.10%

    (4-core case 1.10 pp over the 5% NFR-1 threshold -> observation

    trade-off filed). Cross-platform reading vs the in-project CVM

    A/B (-7.6%~-9.4%) confirms the perf root cause (vendor evolution

    + virtio_user amplification, NOT runtime-fix).

  13.0-baseline-cvm-bench-report.md (514 lines)

    CVM same-timeline A/B (13.0 baseline vs 15.0 runtime-fix-done) +

    Phase 4 perf flame-graph root cause + helloworld_epoll smoke +

    nginx dual-tree A/B + redis dual-tree A/B + bare-metal cross-

    reference (S15). T1/T2/T3 wrk numbers, top-function comparison

    table, differential flame-graph top growers/shrinkers, the 9% gap

    attributed to vendor evolution (TCP stacks vtable / CUBIC / sb_locking

    refactor / virtio path) NOT runtime-fix. Side finding: ext4+inotify

    log noise ~5%.

  runtime-fix-execution-log.md (445 lines)

    Full timeline of the FreeBSD 13.0 -> 15.0 startup-hang debug + fix.

    Carries: 5-role agent team form, DP-DBG-1~3, mandatory conventions

    (rm/kill/chmod scripts + English commit), 6-phase progress, 8-row

    suspect analysis, 6-commit table (5 P0 SIGSEGV fixes + 1 defensive

    panic stub), strict 3-item acceptance evidence, Phase 2 rib-fix

    minimal diff, Phase 3 badfileops kern_descrip boundary fix

    (line 5372 -> 5475 + dropping the {0} stub, with the gdb core-dump

    chain '_fdrop -> kern_accept -> ff_accept -> loop'), Phase 4 perf

    flame-graph reference, helloworld_epoll / nginx / redis dual-tree

    A/B + nginx multi-process functional verification.

Out of scope (intentional, per S2 strategy):

  * config.ini stays modified per the operator's earlier 'F' decision

    (lcore_mask=10 / idle_sleep=20 retained for next replay).

  * M1~M4 execution-log + research-brief, Phase-5b-execution-log,

    plan.md, 98-independent-audit-report.md remain at their current

    English content (delta to zh_cn left for a follow-up task).

Privacy: zero internal IPs (A/B/C segments) leak into any new doc;

the same x.x.x.39 / 192.168.1.X masking convention is reused.

Author: jfb8856606