[feature]: Add reorg-aware block header cache to `LndRpcChainBridge`

[ffranr]

LndRpcChainBridge (lndservices/chain_bridge.go) currently caches only block timestamps by height via blockTimestampCache *lru.Cache[uint32, cacheableTimestamp]. That cache can return timestamps for headers that were reorged out since it is unaware of chain reorgs and confirmation depth. We should replace this with a reorg-aware cache of full block headers.

Goals

• Cache full headers, not just timestamps. Header hash and timestamp are derivable from the header.
• Make GetBlockTimestamp read from the new header cache. Remove or retire blockTimestampCache.
• Implement as a reusable, package-level component with clean interfaces, suitable for use beyond LndRpcChainBridge.

Header cache requirements

• Size is configurable, default target 100k entries.

• On reaching capacity, purge a random fraction (default 10 percent). Fraction is configurable.

• Track confirmation depth. Parameter numberOfConfirmations with typical value 6.

• Store headers even when depth < numberOfConfirmations, but treat them as unsettled:

  • Queries for unsettled heights or hashes return a cache miss to force an external lookup.
  • Once a later settled height is observed that implies sufficient depth, mark prior entries as settled.

• Support lookup by height and by header hash.

• Be resilient to reorgs:

  • Detect conflicting headers at the same height.
  • On detection, invalidate the conflicting branch entries that are now orphaned.

Testing

• Unit tests that cover:

  • Basic put and get by height and by hash.
  • Capacity and random purge behavior.
  • Unsettled vs settled semantics with numberOfConfirmations.
  • Reorg scenario: insert headers for H..H+N, then replace H+k with an alternate header and ensure orphaned entries are invalidated and queries miss until settled.
  • GetBlockTimestamp integration against the new cache.

[ffranr]

PR has been merged into 0-8-0-staging