modules/cuda: completely revamp microarchitecture handling

Replace the per-CUDA-version architecture lists with a declarative
table of support windows (min/max CUDA version per micro-ISA). There's
a clean separation between version bounds and logic, and once a new
arch like Rubin comes out, it's a table entry instead of a change to
complex intertwined logic.

User-visible changes:
- support the specifiers introduced with CUDA 12.8/12.9:
  architecture-specific 'X.Ya' ('Hopper(A)', '12.0a', ...) and
  family-specific 'X.Yf' ('10.0f', ...), plus the Thor and newer
  Blackwell parts
- min_driver_version() table extended through CUDA 13.3
- '+PTX' is now rejected for architecture-specific ('a') targets,
  which have no forward compatibility
- warnings distinguish "CUDA too old for arch" from "arch dropped
  from CUDA"

Author: SoapGentoo

.flake8

@@ -30,4 +30,8 @@ extend-ignore =
     A003
 per-file-ignores =
     mesonbuild/mesonlib.py:F401,F403
+    # E131: continuation line unaligned for hanging indent
+    # E201: whitespace after '('
+    # (both due to deliberate column alignment in the CUDA arch table)
+    mesonbuild/modules/cuda.py:E131,E201
 max-line-length = 120

docs/markdown/Cuda-module.md

@@ -75,20 +75,36 @@ interpretation is:
 The supported architecture names and their corresponding compute capabilities
 are:
 
-| Name              | Compute Capability |
-|-------------------|--------------------|
-| `'Fermi'`         | 2.0, 2.1(2.0)      |
-| `'Kepler'`        | 3.0, 3.5           |
-| `'Kepler+Tegra'`  | 3.2                |
-| `'Kepler+Tesla'`  | 3.7                |
-| `'Maxwell'`       | 5.0, 5.2           |
-| `'Maxwell+Tegra'` | 5.3                |
-| `'Pascal'`        | 6.0, 6.1           |
-| `'Pascal+Tegra'`  | 6.2                |
-| `'Volta'`         | 7.0                |
-| `'Xavier'`        | 7.2                |
-| `'Turing'`        | 7.5                |
-| `'Ampere'`        | 8.0, 8.6           |
+| Name              | Compute Capability         |
+|-------------------|----------------------------|
+| `'Fermi'`         | 2.0, 2.1(2.0)              |
+| `'Kepler'`        | 3.0, 3.5                   |
+| `'Kepler+Tegra'`  | 3.2                        |
+| `'Kepler+Tesla'`  | 3.7                        |
+| `'Maxwell'`       | 5.0, 5.2                   |
+| `'Maxwell+Tegra'` | 5.3                        |
+| `'Pascal'`        | 6.0, 6.1                   |
+| `'Pascal+Tegra'`  | 6.2                        |
+| `'Volta'`         | 7.0                        |
+| `'Xavier'`        | 7.2                        |
+| `'Turing'`        | 7.5                        |
+| `'Ampere'`        | 8.0, 8.6                   |
+| `'Orin'`          | 8.7                        |
+| `'Lovelace'`      | 8.9                        |
+| `'Hopper'`        | 9.0                        |
+| `'Hopper(A)'`     | 9.0a                       |
+| `'Thor'`          | 10.1, 11.0                 |
+| `'Thor(A)'`       | 10.1a, 11.0a               |
+| `'Blackwell'`     | 10.0, 10.3, 12.0, 12.1     |
+| `'Blackwell(A)'`  | 10.0a, 10.3a, 12.0a, 12.1a |
+
+The `'(A)'` names select the architecture-specific (`a`-suffixed) compute
+capabilities of their family. Code built for these can use
+architecture-specific features, but runs only on exactly that architecture:
+it has no forward compatibility, so it cannot be combined with `+PTX`.
+The family-specific `f`-suffixed compute capabilities introduced with
+CUDA 12.9 (e.g. `10.0f`, which runs on all later members of the same
+family) may be requested explicitly, but are not part of any named set.
 
 
 Examples: