feat(interaction): reassemble chunked list-attribute reports for cluster read

[p0fi]

Stacked on #100

This PR is stacked on #100 (tlv.Optional[T]) and targets that branch instead of main. Once #100 merges, GitHub will retarget this PR to main automatically.

Summary

• Devices split a list attribute too large for one AttributeDataIB into an empty/partial-array marker followed by ListIndex=null "append" fragments; cluster read previously surfaced these as separate rows instead of one reassembled value.
• Merges fragments into one AttributeReport per logical attribute inside internal/interaction.Client.Read, after its existing MoreChunkedMessages chunk-collection loop, operating on raw TLV bytes (no dependency on the clusters package).
• Fragment contiguity (one attribute's fragments are never interleaved with another path's reports, even across message boundaries) was verified against the C++ reference implementation's reporting engine (Engine::BuildSingleReportDataAttributeReportIBs / AttributeEncodeState in connectedhomeip) before writing the merge algorithm.
• A malformed sequence (an append fragment with no preceding whole-list marker) passes through as a single unmerged row rather than failing the whole read.
• cli/cluster.go and internal/daemon are untouched — both already produce one row per report, so fewer, fully-merged reports from Client.Read fix cluster read for free.

Closes #96

Test plan

• mise run test (table-driven tests in internal/interaction/reassemble_test.go with hand-built TLV fixtures, including a malformed-sequence case)
• mise run lint
• mise run build
• Two-axis /code-review (Standards + Spec) run against this diff — no hard findings on either axis

🤖 Generated with Claude Code