TOML encoder: prefer readable table sections over inline tables

[Copilot]

When encoding to TOML from YAML/JSON, nested mappings were collapsed into compact inline table syntax (key = { ... }) instead of idiomatic TOML table sections ([key]). This affected both direct YAML→TOML conversion and round-trips through YAML.

Changes

Encoder (encoder_toml.go)

• encodeTopLevelEntry: Replace EncodeSeparate check with FlowStyle check — only nodes explicitly marked as flow/inline stay inline; all block mappings become table sections
• encodeSeparateMapping: Count FlowStyle mapping children as hasAttrs (they render as key = { ... }); replace writeTableHeader + encodeMappingBodyWithPath with recursive encodeSeparateMapping in the "no attrs" loop
• encodeMappingBodyWithPath: Emit non-flow child mappings as sub-table sections ([parent.child]) instead of inline tables

Decoder (decoder_toml.go)

• Set Style: FlowStyle on decoded inline TOML tables, distinguishing them from block mappings so they survive re-encoding as inline

Tests (toml_test.go)

• Add encode scenario type (YAML→TOML) and coverage for table section output
• Update roundtrip expectations: inline TOML tables now expand to table sections (more readable)

Before / After

# input YAML
arg:
  hello: foo

# before
arg = { hello = "foo" }

# after
[arg]
hello = "foo"

YAML flow mappings (arg: {hello: foo}) are preserved as TOML inline tables, respecting the author's explicit style intent.