Preserve HTTP form encoder configuration

[Gimini-3]

Fixes #4330

What changed

• Preserve pending charset, encoding mode, multipart, factory, and cleanup configuration across encoder rebuilds.
• Apply pending encoder changes before adding form parts so a custom HttpDataFactory actually owns the created data.
• Route later builder calls and cleanup to the latest encoder, including callback-failure cleanup.
• Align the cleanOnTerminate Javadoc with its actual default.

Failing regression coverage

Before the fix, the new tests demonstrate two observable setter-order failures:

charset(ISO_8859_1).multipart(false): key=%C3%A9
expected:                               key=%E9

encoding(RFC3986).multipart(false): key=*
expected:                          key=%2A

They also verify that cleanOnTerminate(false) survives charset, encoding, multipart, and factory-triggered rebuilds, and that both normal and callback-failure paths clean the configured custom factory.

Scope

The production change is limited to HTTP client form configuration/rebuild and error cleanup:

• HttpClientForm
• HttpClientFormEncoder
• HttpClientOperations.SendForm

Tests cover both the encoder directly and real HTTP form uploads.

Validation

• HttpClientFormEncoderTests: 9 passed
• HttpClientWithTomcatTest: 15 passed
• checkstyleMain, checkstyleTest, and spotlessCheck: passed
• Full reactor-netty-http suite: 2,888 tests executed; one unrelated SNI handshake-timeout test failed once and passed on isolated rerun