allow forcing full (non-range) HTTP downloads for servers with non-compliant Range/206 responses with SET force_download = true

Some public APIs, e.g. France's [hubeau.eaufrance.fr](https://hubeau.eaufrance.fr/) (data.gouv.fr), return non-RFC7233-compliant Range responses (206 status with mismatched Content-Length/Content-Range). httpfs treats this as fatal instead of falling back to a plain download.

### To Reproduce
```
FROM read_json('https://hubeau.eaufrance.fr/api/v1/qualite_eau_potable/resultats_dis?code_commune=31555');
Error: HTTP Error: Full download failed to URL "...": 206 (Partial Content)
```
Same root cause was already reported 2+ years ago in [#11221](https://github.com/duckdb/duckdb/issues/11221), closed as stale with no fix or workaround shipped. Still reproduces today. I've contacted Hubeau directly; a header-compliance fix on their side isn't realistic on any useful timeframe (public-administration API, limited bandwidth).

This isn't one quirky server: hubeau.eaufrance.fr is one of France's most-used open-data platforms, covering water data (drinking water quality, river quality, groundwater levels, hydrometry, etc.).

### Proposed solution
A session-level setting to force a full sequential GET, skipping HEAD/Range logic and Content-Length/Content-Range validation, with the common force_download setting, so have it work with this setting (it doesn't currently):

```
SET force_download = true;
FROM read_json('https://hubeau.eaufrance.fr/api/v1/qualite_eau_potable/resultats_dis?code_commune=31555');
```
### Alternatives considered
Pre-downloading with curl or Python (requests + temp file) works, but defeats the point of httpfs and isn't viable in SQL-only contexts (CLI, dbt, BI tools embedding DuckDB).

The issue names no implementation files or tests. Start by tracing the force_download setting and the read_json HTTP path in httpfs, then inspect handling of HEAD, Range, and 206 responses; done means the Hubeau reproduction works with SET force_download = true without the current validation failure.

Written by the indexing model from the issue text.