"bulk" requests failing due to multiple conflicting Accept headers

**Describe the bug/error/problem**
`bulk` requests from the v9 client to a v9 server can fail with a 400 response caused by a duplicate `Accept` header.

Using tcpdump, I observed these headers being sent:

```
Accept: */*
Content-Type: application/vnd.elasticsearch+x-ndjson; compatible-with=9
accept: application/vnd.elasticsearch+x-ndjson; compatible-with=9
```

And this was the error log from fluentd:

`
[warn]: #0 failed to flush the buffer. retry_times=0 next_retry_time=2026-08-28 12:21:24 +0000 chunk="65a1a7d518cf9ca154f2901a49baed5b" error_class=Fluent::Plugin::ElasticsearchOutput::RecoverableRequestFailure error="could not push logs to Elasticsearch cluster ({host: \"127.0.0.1\", port: 9200, scheme: \"http\"}): [400] {\"error\":{\"root_cause\":[{\"type\":\"media_type_header_exception\",\"reason\":\"Invalid media-type value on headers [Accept]\"}],\"type\":\"media_type_header_exception\",\"reason\":\"Invalid media-type value on headers [Accept]\",\"caused_by\":{\"type\":\"illegal_argument_exception\",\"reason\":\"Incorrect header [Accept]. Only one value should be provided\"},\"suppressed\":[{\"type\":\"media_type_header_exception\",\"reason\":\"Invalid media-type value on headers [Accept, Content-Type]\",\"caused_by\":{\"type\":\"status_exception\",\"reason\":\"A compatible version is required on both Content-Type and Accept headers if either one has requested a compatible version. Accept=null, Content-Type=application/vnd.elasticsearch+x-ndjson; compatible-with=9\"}}]},\"status\":400}"
`

In my setup, I have fluentd using fluent-plugin-elasticsearch and the elasticsearch client gem to send logs to Elasticsearch. I experienced the problem on version 9.5.0 of the client; downgrading to 8.19.3 solved it.

**Your Environment (please complete the following information):**
- Operating System: RHEL UBI 9
- Ruby Version: 4.0.6 from RHEL AppStream
- Elasticsearch client version: 9.5.0
- Elasticsearch version: 9.5.2
- Any other relevant information:
- used with fluentd 1.19.3 and fluent-plugin-elasticsearch 6.0.0
- faraday 2.14.3
- excon 1.7.0

**Additional context**
I used an AI agent to help me investigate this. It pointed out that #2759 writes the headers as lowercase -- `accept` -- and suggests that the Faraday/Excon HTTP stack isn't smart enough to recognise that, and just adds the uppercase version. I haven't traced that through personally though.

Start by tracing how the v9 client constructs headers for bulk requests and how the Faraday/Excon stack merges them, using the duplicate headers and #2759 as the initial clues. Reproduce the request against Elasticsearch 9.5.2 with the listed client versions; done means a bulk request sends only one compatible Accept header and no longer receives the 400 response.

Written by the indexing model from the issue text.