Updated CHANGELOG for v1.4.7

[Cyan4973]

@felixhandte , @terrelln , @senhuang42 ,
please have a look to make sure that all topics you want to be mentioned are effectively present.

A few additional notes :

• the next release is presumed to be numbered v1.4.7.
  This will automatically bring the question "what about v1.4.6 ?".
  I think we can mention that v1.4.6 is a non-public release.
  If that feels too much hassle, we could also decide to call this coming release v1.4.6,
  though we will then need another discriminant to differentiate from the kernel release.

  • Note that the next action is to update source code and documentation to v1.4.7 if this is the preferred choice
    (likely a follow-up commit in this PR).

• My bullet-point list (v1.4.7) is not as dense as the v1.4.6 one.
  I only care to mention changes which are visible to users (perf, cli, api, package, etc.)
  while invisible stuff (improved CI and tests, internal source reorg, etc) doesn't make it to CHANGELOG.
  We could keep it this way, or decide to unify policy for v1.4.6 and v1.4.7, or even fuse these 2 paragraphs if that feels preferable.

[ghost]

It's worthy to mention issue #2327: fix multithreaded output can depend on number of threads, by @terrelln

And PR #2303 by @senhuang42 :

1. When using ZSTD_compressStream2(), if a cdict was used during compression, that cdict's compression level was not respected. Now the cdict will take priority in terms of compression level, but will not change cctx->requestedParams.
2. Fixes ZSTD_createCDict_byReference() to use the overloaded compressionLevel = 0 behavior.

[ghost]

api : fix : multithreaded compression has predictable output even in special cases (see #2327) (issue not accessible from cli)

#2327 is the issue's number, it seems all numbers in CHANGLOG are PR numbers.
The PR number of this problem is #2339.

[senhuang42]

Might be worth mentioning the Makefile build changes? Incremental builds and config detection are pretty nice user-facing improvements.