[fpga] docs: record Toy/Linux hardware bring-up and programming entry

[Cody-ic]

Record the first physical Arty A7-100T Toy/Linux boot from PR #5 and provide a discoverable program.tcl entry for repeating JTAG programming. The platform guides now report the successful hardware result while preserving the original October 11 implementation/RTL evidence and remaining acceptance limits.

Changes

• Archive the 14 original programming/serial/result/procedure artifacts with unchanged bytes and explicit SHA256 entries, including the actual program_arty.tcl used on the board.
• Document complete DDR image readback, Linux boot, five PASS checks, live shell, the expected page-protection SIGSEGV and the preserved unsuccessful auxiliary shell probe.
• Add platform-root program.tcl, delegating to the existing measured board script. The PowerShell wrapper resolves the bitstream before changing directories, supporting relative paths and paths with spaces.
• Check the physical archive in the existing CI source verifier; update build/programming instructions and volatile SRAM/DDR behavior.

Validation

• Physical JTAG programming: XC7A100T, Vivado 2025.1, startup status HIGH.
• All 17,409,032 bytes of the original payload written and individually read back before RUN; 544.1 seconds. Original bitstream/payload hashes match the validated implementation.
• Physical OpenSBI v1.9/Linux 6.6.0/BusyBox boot. COW, page protection, FP64 process context, timer and total system check PASS; process return code 0. Subsequent /proc/version and /proc/meminfo shell reads pass.
• 205 source/evidence hashes and all 14 original hardware artifacts verified. Offline Tcl smoke check covers entry-directory independence, spaced paths and argument/file/target/device rejection; PowerShell syntax and Python compilation pass. Gitleaks 8.30.1 reports no leaks.
• Hardware RTL, constraints, synthesis script, measured board programming script and original host loader are unchanged. The new Tcl entry was verified offline; packaging did not reprogram or disturb the running Linux board.

Scope

PCB revision remains unconfirmed; the exact E.0 MIG engineering image works on the connected board. Physical Gemmini regression, exhaustive FP16/FP32 checks, full DDR sweep, long-duration stability and power tests remain pending. Slice occupancy remains 98.82%. Build products are not committed: users build/select their .bit and Linux payload, then use the documented Tcl and UART commands.