docs: clarify RISC-V RVV target selection and GCC 14+ requirement for ZVL128B/ZVL256B

Add two notes to the RISC-V section before the per-target entries:

1. RISCV64_GENERIC is intentionally scalar — Makefile.riscv64 appends a
   scalar march override that takes precedence over any user-supplied
   -march=rv64gcv. Correct targets for RVV 1.0 are RISCV64_ZVL128B and
   RISCV64_ZVL256B (see #3808 for design rationale).

2. GCC 14+ required for _rvv.c kernels on current OpenBLAS. GCC 13
   builds complete and produce a library but routines using segmented
   load/store intrinsics (__riscv_vsseg*) fall back to scalar silently.
   Functional tests pass; only disassembly detects this. Verified on
   OpenBLAS 0.3.33: GCC 13 (scalar fallback), GCC 14 (~12,691 RVV
   opcodes), GCC 15 (~14,355 RVV opcodes).

Co-authored-by: Vaibhav805 <Vaibhav805@users.noreply.github.com>

Author: trg-rgb

README.md

@@ -226,6 +226,10 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
   make HOSTCC=gcc TARGET=x280 NUM_THREADS=8 CC=riscv64-unknown-linux-gnu-clang FC=riscv64-unknown-linux-gnu-gfortran
   ```
 
+**Note — RVV 1.0 target selection:** `RISCV64_GENERIC` is intentionally scalar — its kernel file maps all BLAS operations to plain-C implementations, and `Makefile.riscv64` appends a scalar `-march` override for this target that takes precedence over any user-supplied `-march=rv64gcv` flag. For RVV 1.0 vectorized builds, use `TARGET=RISCV64_ZVL128B` (VLEN ≥ 128 bits) or `TARGET=RISCV64_ZVL256B` (VLEN ≥ 256 bits). These targets route all three BLAS levels — including DGEMM — to the `_rvv.c` kernel set introduced in 2022; see [issue #3808](https://github.com/OpenMathLib/OpenBLAS/issues/3808) for the design rationale.
+
+**Compiler requirement for ZVL targets:** GCC 14 or later is required on current OpenBLAS releases. GCC 13 does not implement the segmented load/store intrinsics (`__riscv_vsseg*`) used by the `_rvv.c` kernels; the build still completes and produces a library, but the affected routines fall back to scalar code paths. Functional tests will pass on the resulting library — only disassembly-level verification detects the regression. For a correct `RISCV64_ZVL128B` build on OpenBLAS 0.3.33, `riscv64-linux-gnu-objdump -d libopenblas*.a | grep -c 'vle64\|vfmacc\|vsetvli\|vlse64\|vfmul\|vfadd\|vfredosum'` returns approximately 12,000–14,000 (GCC 14: ~12,691; GCC 15: ~14,355).
+
 - **ZVL???B**: Level-3 BLAS and Level-1,2 including vectorised kernels targeting generic RISCV cores with vector support with registers of at least the corresponding width; ZVL128B and ZVL256B are available.
 e.g.:
   ```sh