Originally, clangd-tidy was enabled
[here](duckdb#21745).

The CI check for `clangd-tidy` was disabled in
[this](duckdb#23196) PR because the CI
job ran very slowly:
<img width="822" height="550" alt="Screenshot 2026-07-29 at 10 40 35"
src="https://github.com/user-attachments/assets/38d540c5-a188-4ec4-8f39-51533df183f0"
/>

The job failed because the script exhausted all disk space (160 GiB!).
The script kept recreating precompiled header files from scratch after a
tidy error, but it did not remove the previously created files.

Fixes duckdblabs/duckdb-internal#9770.

### Changes
- Re-enable the `clangd-tidy` check in CI (for PRs and on main).
- Write `clangd-tidy` disk stats and tidy attempts to CI output.
- Use 8 evenly sized chunks and print failures immediately in CI output.
- Stop after max 3 failed chunks (so at most 3x 2000 error lines, 3x
~200 files).

### Tidy fixes
- Fix `performance-unnecessary-copy-initialization` error (missing
`const auto`).
- Remove ineffective `std::move()` calls.

### Disk usage

The disk usage is kept at 40 GiB (25% of that CI runner instance):

```
attempt 1 (retry 0/2, files 204): tools/utils/test_platform.cpp
attempt 1 finished exit=0 in 38.2s; pch size: 5.8 GiB
attempt 2 (retry 0/2, files 204): src/optimizer/window_self_join.cpp
attempt 2 finished exit=0 in 36.1s; pch size: 11.0 GiB
attempt 3 (retry 0/2, files 203): src/planner/binder/tableref/bind_showref.cpp
attempt 3 finished exit=0 in 29.0s; pch size: 14.8 GiB
attempt 4 (retry 0/2, files 203): src/parser/peg/transformer/transform_copy.cpp
attempt 4 finished exit=0 in 27.3s; pch size: 18.8 GiB
attempt 5 (retry 0/2, files 203): src/function/scalar/system/aggregate_export.cpp
attempt 5 finished exit=0 in 32.0s; pch size: 23.7 GiB
attempt 6 (retry 0/2, files 203): src/common/crypto/md5.cpp
attempt 6 finished exit=0 in 29.9s; pch size: 28.1 GiB
attempt 7 (retry 0/2, files 203): src/execution/operator/helper/physical_disconnect.cpp
attempt 7 finished exit=0 in 38.5s; pch size: 34.3 GiB
attempt 8 (retry 0/2, files 203): src/main/relation/delim_get_relation.cpp
attempt 8 finished exit=0 in 35.7s; pch size: 40.4 GiB
```

### Measurements

Originally, clangd-tidy took ~9 min to run. I did some experiments on
how to use the CI instance (16 vCPU, 32 GiB RAM) more efficiently:

chunks | concurrency | average CPU | duration
--- | --- | --- | ---
100-file chunks | `-j4` | 53% |
[10m17s](https://cloud.namespace.so/rrete04q43id4/actions/job/90520079762)
200-file chunks | `-j4` | 55% |
[6m59s](https://cloud.namespace.so/rrete04q43id4/actions/job/90520079762)
10 evenly sized chunks | `-j8` | 95% |
[4m16s](https://cloud.namespace.so/rrete04q43id4/actions/job/90523783761)
8 evenly sized chunks | `-j8` | 93% |
[4m28s](https://cloud.namespace.so/rrete04q43id4/actions/job/90525308431)

It's better to use fewer (10 -> 8) chunks because an error clears the
precompiled header directory causing clangd to rebuild included header
files from scratch.