> Everybody implementing their abstractions, every library implementing their own language
In Lisp one can extend the language syntax. One does not need to implement a new language. For example the Common Lisp Object System adds operators like DEFCLASS, DEFMETHOD, DEFGENERIC, ... in addition to already existing operators like DEFUN, DEFSTRUCT, DEFTYPE, DEFPARAMETER, ...
Thus the language can be incrementally extended, instead of using a completely new language.
The CLOS macros are used by many programmers, they were standardized, documented and implemented.
As every form of abstraction, syntactic abstraction needs to be learned and requires extra work. Lisp was from the start developed to enable such things, and its early form of that applied to Lisp itself were the Lisp interpreter evaluating s-expressions, FEXPRs (procedured receiving unevaluated arguments) and in 1962 macros.
> I feel like Lisp and Haskell attract people that love programming more than shipping code.
One of the purposes of Lisp was to implement new ideas: like computing with mathematical formulas (-> Macsyma and earlier attempts), computing theorems (-> ACL2 and earlier attempts), ... This ability to support experimentation in R&D (actors, rule-based systems, frame systems, ...) was always a part of the language community.
If we look at the surviving commercial systems (Allegro CL and LispWorks), they still support this feature set.
Clojure was developed as a Lisp-inspired functional language with deep Java/JVM integration to support enterprise programming for people who were tired of programming in Java. That's fine. But the advice then was to not utilize the full power of Lisp (syntactic abstractions), to address the fear of unmaintainable software, while at the same time not using various features of Lisp (like gradually typed implementations of Lisp like SBCL) which helps to write robust software.
Generally Lisp has a broader outlook. It ran on an experimental spacecraft, it powered autonomous robots inspecting pipelines, it was the base for early research in computing (-> Interlisp), it ran on calculators (HP RPL), it supported children education (Logo), it's used to schedule telescope operations (-> Hubble and James Webb), it is used to schedule airline/airport operations, ...
> php has more killer software than all those languages combined.
If we research for ground breaking software, then I bet over the long history of Lisp there were a lot more in many more diverse domains written in Lisp
Lisp has been applied to develop airplane parts. BOEING and Airbus used to be Lisp users (they might be still use it for older planes). ICAD was the first parametric CAD system. ICAD used macros to describe physical objects and their relationships. This was used to construct turbines, wings, and a lot of other parts of Airplanes.
If you look into the long history of Lisp, it has been successfully applied in domains PHP has never seen. Sometimes really crazy stuff, like the first virtual military troop training systems, where the US army deployed a few extensive training and simulation systems, where the virtual worlds were controlled & generated by a Lisp system and rendered by multiple graphics engines for each training simulator. In the early 80s. -> https://en.wikipedia.org/wiki/SIMNET
That's nothing like PHP, but it was ground breaking in its domain. PHP had a different purpose: it was developed to make it simple to write&generate web pages in a large scale. That's a huge domain, but that was not what Lisp was developed for.
Btw., we see a bunch of new languages which now also support some form of macros. A recent example is Rust.